├── .gitignore ├── A-0004-01_P.pdf ├── AffineInvariantFeatures.py ├── LICENSE ├── README.md ├── RegionOfInterest.py ├── TemplateMatcher.py ├── avery_8254.pdf ├── classifier.pdf ├── classifier.pkl ├── classifier_alt_1.pkl ├── classifier_alt_2.pkl ├── classify.py ├── dns_badge.png ├── dns_logo.png ├── sample.jpg ├── setup.py ├── template.png ├── train.py └── util.py /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | env/ 12 | build/ 13 | develop-eggs/ 14 | dist/ 15 | downloads/ 16 | eggs/ 17 | .eggs/ 18 | lib/ 19 | lib64/ 20 | parts/ 21 | sdist/ 22 | var/ 23 | *.egg-info/ 24 | .installed.cfg 25 | *.egg 26 | 27 | # PyInstaller 28 | # Usually these files are written by a python script from a template 29 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 30 | *.manifest 31 | *.spec 32 | 33 | # Installer logs 34 | pip-log.txt 35 | pip-delete-this-directory.txt 36 | 37 | # Unit test / coverage reports 38 | htmlcov/ 39 | .tox/ 40 | .coverage 41 | .coverage.* 42 | .cache 43 | nosetests.xml 44 | coverage.xml 45 | *,cover 46 | .hypothesis/ 47 | 48 | # Translations 49 | *.mo 50 | *.pot 51 | 52 | # Django stuff: 53 | *.log 54 | local_settings.py 55 | 56 | # Flask stuff: 57 | instance/ 58 | .webassets-cache 59 | 60 | # Scrapy stuff: 61 | .scrapy 62 | 63 | # Sphinx documentation 64 | docs/_build/ 65 | 66 | # PyBuilder 67 | target/ 68 | 69 | # IPython Notebook 70 | .ipynb_checkpoints 71 | 72 | # pyenv 73 | .python-version 74 | 75 | # celery beat schedule file 76 | celerybeat-schedule 77 | 78 | # dotenv 79 | .env 80 | 81 | # virtualenv 82 | venv/ 83 | ENV/ 84 | 85 | # Spyder project settings 86 | .spyderproject 87 | 88 | # Rope project settings 89 | .ropeproject 90 | -------------------------------------------------------------------------------- /A-0004-01_P.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/AVGInnovationLabs/DoNotSnap/e679947fd3bdfec7c9c786ee285e9ac64d8d1a18/A-0004-01_P.pdf -------------------------------------------------------------------------------- /AffineInvariantFeatures.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import numpy as np 3 | import itertools as it 4 | from multiprocessing.pool import ThreadPool 5 | from sklearn.base import BaseEstimator, TransformerMixin 6 | 7 | 8 | class AffineInvariant(TransformerMixin, BaseEstimator): 9 | def __init__(self, detector, extractor): 10 | self.detector = detector 11 | self.extractor = extractor 12 | self.pool = ThreadPool(processes=cv2.getNumberOfCPUs()) 13 | 14 | def affine_skew(self, tilt, phi, img, mask=None): 15 | h, w = img.shape[:2] 16 | if mask is None: 17 | mask = np.zeros((h, w), np.uint8) 18 | mask[:] = 255 19 | A = np.float32([[1, 0, 0], [0, 1, 0]]) 20 | if phi != 0.0: 21 | phi = np.deg2rad(phi) 22 | s, c = np.sin(phi), np.cos(phi) 23 | A = np.float32([[c, -s], [s, c]]) 24 | corners = [[0, 0], [w, 0], [w, h], [0, h]] 25 | tcorners = np.int32(np.dot(corners, A.T)) 26 | x, y, w, h = cv2.boundingRect(tcorners.reshape(1, -1, 2)) 27 | A = np.hstack([A, [[-x], [-y]]]) 28 | img = cv2.warpAffine(img, A, (w, h), flags=cv2.INTER_LINEAR, borderMode=cv2.BORDER_REPLICATE) 29 | if tilt != 1.0: 30 | s = 0.8*np.sqrt(tilt * tilt - 1) 31 | img = cv2.GaussianBlur(img, (0, 0), sigmaX=s, sigmaY=0.01) 32 | img = cv2.resize(img, (0, 0), fx=1.0 / tilt, fy=1.0, interpolation=cv2.INTER_NEAREST) 33 | A[0] /= tilt 34 | if phi != 0.0 or tilt != 1.0: 35 | h, w = img.shape[:2] 36 | mask = cv2.warpAffine(mask, A, (w, h), flags=cv2.INTER_NEAREST) 37 | Ai = cv2.invertAffineTransform(A) 38 | return img, mask, Ai 39 | 40 | def affine_detect(self, img, mask=None): 41 | params = [(1.0, 0.0)] 42 | for t in 2 ** (0.5 * np.arange(1, 6)): 43 | for phi in np.arange(0, 180, 72.0 / t): 44 | params.append((t, phi)) 45 | 46 | def f(p): 47 | t, phi = p 48 | timg, tmask, Ai = self.affine_skew(t, phi, img) 49 | keypoints = self.detector.detect(timg, tmask) 50 | keypoints, descrs = self.extractor.compute(timg, keypoints) 51 | for kp in keypoints: 52 | x, y = kp.pt 53 | kp.pt = tuple(np.dot(Ai, (x, y, 1))) 54 | if descrs is None: 55 | descrs = np.zeros((0, 64), 'float32') 56 | return keypoints, descrs 57 | 58 | keypoints, descrs = [], [] 59 | ires = it.imap(f, params) 60 | 61 | for i, (k, d) in enumerate(ires): 62 | keypoints.extend(k) 63 | descrs.extend(d) 64 | 65 | return keypoints, np.array(descrs) 66 | 67 | def extract_features(self, image): 68 | keypoints, descriptors = self.affine_detect(image) 69 | if descriptors is None or not len(descriptors): 70 | # treat failure to classify as negative sample 71 | return np.zeros((0, 64), 'float32') 72 | 73 | return descriptors 74 | 75 | def fit(self, X, y=None, **fit_params): 76 | return self 77 | 78 | def transform(self, X, **transform_params): 79 | l = dict(n=0) 80 | 81 | def f(image): 82 | l['n'] += 1 83 | if (100 * l['n'] / len(X)) % 5 == 0: 84 | print '\r%d%% - %d/%d' % (100 * l['n'] / len(X), l['n'], len(X)), 85 | return self.extract_features(image) 86 | 87 | ires = self.pool.imap(f, X) 88 | return np.array(list(ires)) 89 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | {one line to give the program's name and a brief idea of what it does.} 635 | Copyright (C) {year} {name of author} 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | {project} Copyright (C) {year} {fullname} 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # DoNotSnap 2 | 3 | 4 | An experiment in detecting DoNotSnap badges in photos, to protect privacy. 5 | 6 |

7 | 8 | This program allows you to detect and identify DoNotSnap badges via a sliding-window decision tree classifier (custom heuristics are used to reduce search space). 9 | The classifier is trained by matching samples against image templates using Affine-transform invariant SURF features. 10 | 11 | You can find examples of using the classifier in `classify.py` and training a new classifier in `train.py` 12 | 13 | A pre-trained classifier can be found in `classifier.pkl` 14 | Alternative versions of the same classifier are in `classifier_alt_1.pkl` and `classifier_alt_2.pkl` 15 | ## Running classification 16 | Run `python classify.py ` 17 | This will deserialize the classifier from `classifier.pkl` and run it on the image you supplied. A sample image could be found in `sample.jpg` 18 | ## Training your own classifier 19 | Run `python train.py ` 20 | This will read the sample filenames from `positive.txt` and `negative.txt` files. 21 | Templates filenames are specified in `templates.txt`. A sample template could be found in `template.png` 22 | The output is a `.pkl` with serialized classifier. 23 | ## Dependencies 24 | * opencv 25 | * numpy 26 | * sklearn 27 | * matplotlib 28 | * PIL 29 | 30 | ## DoNotSnap sticker templates 31 | In the repository you can find templates for printing DoNoSnap stickers. 32 | 33 | `avery_8254.pdf` contains design compatible with US format paper. Compatible Avery templates: 34 | * 15664 35 | * 18664 36 | * 45464 37 | * 48264 38 | * 48464 39 | * 48864 40 | * 5164 41 | * 5264 42 | * 55164 43 | * 5524 44 | * 55264 45 | * 55364 46 | * 55464 47 | * 5664 48 | * 58164 49 | * 58264 50 | * 8164 51 | * 8254 52 | * 8464 53 | * 8564 54 | * 15264 55 | * 95940 56 | * 95905 57 | 58 | `A-0004-01_P.pdf` contains design compatible with A4-format paper. Compatible Avery templates: 59 | * J8165 60 | * J8165-10 61 | * J8165-25 62 | * J8165-40 63 | * J8165-100 64 | * J8165-250 65 | * J8165-500 66 | * J8365 67 | * J8565 68 | * J8565-25 69 | * L7165 70 | * L7165-40 71 | * L7165-100 72 | * L7165-250 73 | * L7165-500 74 | * LR7165-100 75 | * L7165X 76 | * L7165X-100 77 | * L7165X-250 78 | * L7565 79 | * L7565-25 80 | * L7965 81 | * L7965-100 82 | * L7993 83 | * L7993-25 84 | * LR7165 85 | * LR7165-100 86 | -------------------------------------------------------------------------------- /RegionOfInterest.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import math 3 | import numpy as np 4 | from util import pyramid, sliding_window 5 | 6 | 7 | def boundingRects(scale, contours): 8 | for contour in contours: 9 | epsilon = 0.1 * cv2.arcLength(contour, True) 10 | approx = cv2.approxPolyDP(contour, epsilon, True) 11 | x, y, w, h = cv2.boundingRect(approx) 12 | 13 | yield [x * scale, y * scale, w * scale, h * scale] 14 | 15 | 16 | def extractEdges(hue, intensity): 17 | kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3)) 18 | 19 | edges = cv2.Canny(intensity, 120, 140) 20 | hue_edges = cv2.Canny(cv2.GaussianBlur(hue, (5, 5), 0), 0, 255) 21 | combined_edges = cv2.bitwise_or(hue_edges, edges) 22 | _, mask = cv2.threshold(combined_edges, 40, 255, cv2.THRESH_BINARY) 23 | return cv2.erode(cv2.GaussianBlur(mask, (3, 3), 0), kernel, iterations=1) 24 | 25 | 26 | def roiFromEdges(edges): 27 | kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (7, 7)) 28 | small_kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3)) 29 | # close gaps in edges to create continous regions 30 | roi = cv2.dilate(edges, small_kernel, iterations=14) 31 | return cv2.erode(roi, kernel, iterations=4) 32 | 33 | 34 | def findEllipses(edges): 35 | contours, _ = cv2.findContours(edges.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE) 36 | ellipseMask = np.zeros(edges.shape, dtype=np.uint8) 37 | contourMask = np.zeros(edges.shape, dtype=np.uint8) 38 | 39 | pi_4 = np.pi * 4 40 | 41 | for i, contour in enumerate(contours): 42 | if len(contour) < 5: 43 | continue 44 | 45 | area = cv2.contourArea(contour) 46 | if area <= 100: # skip ellipses smaller then 10x10 47 | continue 48 | 49 | arclen = cv2.arcLength(contour, True) 50 | circularity = (pi_4 * area) / (arclen * arclen) 51 | ellipse = cv2.fitEllipse(contour) 52 | poly = cv2.ellipse2Poly((int(ellipse[0][0]), int(ellipse[0][1])), (int(ellipse[1][0] / 2), int(ellipse[1][1] / 2)), int(ellipse[2]), 0, 360, 5) 53 | 54 | # if contour is circular enough 55 | if circularity > 0.6: 56 | cv2.fillPoly(ellipseMask, [poly], 255) 57 | continue 58 | 59 | # if contour has enough similarity to an ellipse 60 | similarity = cv2.matchShapes(poly.reshape((poly.shape[0], 1, poly.shape[1])), contour, cv2.cv.CV_CONTOURS_MATCH_I2, 0) 61 | if similarity <= 0.2: 62 | cv2.fillPoly(contourMask, [poly], 255) 63 | 64 | return ellipseMask, contourMask 65 | 66 | 67 | def findCircles(hue, intensity): 68 | houghCirclesMask = np.zeros(hue.shape, dtype=np.uint8) 69 | 70 | blurred_hue = cv2.GaussianBlur(hue, (9, 9), 2) 71 | blurred_intensity = cv2.GaussianBlur(intensity, (9, 9), 2) 72 | hue_circles = cv2.HoughCircles(blurred_hue, cv2.cv.CV_HOUGH_GRADIENT, 0.5, hue.shape[0] / 8, param1=10, param2=25, maxRadius=100) 73 | intensity_circles = cv2.HoughCircles(blurred_intensity, cv2.cv.CV_HOUGH_GRADIENT, 0.5, hue.shape[0] / 8, param1=185, param2=20, maxRadius=100) 74 | 75 | circles = np.vstack((hue_circles[0] if hue_circles is not None else np.empty((0, 3)), 76 | intensity_circles[0] if intensity_circles is not None else np.empty((0, 3)))) 77 | 78 | for (x, y, r) in circles: 79 | cv2.circle(houghCirclesMask, (int(round(x)), int(round(y))), int(round(r)), 255, -1) 80 | 81 | return houghCirclesMask 82 | 83 | 84 | def findRANSACCircles(edges, circleSearches=5): 85 | edges = edges.copy() 86 | mask = np.zeros(edges.shape, dtype=np.uint8) 87 | 88 | minRadius = 10 89 | maxRadius = 100 90 | 91 | def verifyCircle(dt, center, radius): 92 | minInlierDist = 2.0 93 | maxInlierDistMax = 100.0 94 | maxInlierDist = max(minInlierDist, min(maxInlierDistMax, radius / 25.0)) 95 | 96 | # choose samples along the circle and count inlier percentage 97 | samples = np.arange(0, 2 * np.pi, 0.05) 98 | cX = radius * np.cos(samples) + center[0] 99 | cY = radius * np.sin(samples) + center[1] 100 | 101 | coords = np.array((cX, cY)).T 102 | counter = len(samples) 103 | 104 | cXMask = (cX < dt.shape[1]) & (cX >= 0) 105 | cYMask = (cY < dt.shape[0]) & (cY >= 0) 106 | cMask = cXMask & cYMask 107 | 108 | gdt = dt[cY[cMask].astype(int), cX[cMask].astype(int)] 109 | dtMask = gdt < maxInlierDist 110 | 111 | inlierSet = coords[cMask][dtMask] 112 | inlier = len(inlierSet) 113 | 114 | return float(inlier) / counter, inlierSet 115 | 116 | def getCircle(p1, p2, p3): 117 | x1 = float(p1[0]) 118 | x2 = float(p2[0]) 119 | x3 = float(p3[0]) 120 | 121 | y1 = float(p1[1]) 122 | y2 = float(p2[1]) 123 | y3 = float(p3[1]) 124 | 125 | center_x = (x1 * x1 + y1 * y1) * (y2 - y3) + (x2 * x2 + y2 * y2) * (y3 - y1) + (x3 * x3 + y3 * y3) * (y1 - y2) 126 | x = 2 * (x1 * (y2 - y3) - y1 * (x2 - x3) + x2 * y3 - x3 * y2) 127 | if not x: 128 | return None, None 129 | center_x /= x 130 | 131 | center_y = (x1 * x1 + y1 * y1) * (x3 - x2) + (x2 * x2 + y2 * y2) * (x1 - x3) + (x3 * x3 + y3 * y3) * (x2 - x1) 132 | y = 2 * (x1 * (y2 - y3) - y1 * (x2 - x3) + x2 * y3 - x3 * y2) 133 | if not y: 134 | return None, None 135 | center_y /= y 136 | 137 | radius = math.sqrt((center_x - x1) * (center_x - x1) + (center_y - y1) * (center_y - y1)) 138 | 139 | return (center_x, center_y), radius 140 | 141 | def getPointPositions(binaryImage): 142 | return [(x, y) for y, x in zip(*np.where(binaryImage > 0))] 143 | 144 | for _ in xrange(circleSearches): 145 | edgePositions = getPointPositions(edges) 146 | 147 | # create distance transform to efficiently evaluate distance to nearest edge 148 | dt = cv2.distanceTransform(255 - edges, cv2.cv.CV_DIST_L1, 3) 149 | 150 | bestCircleCenter = None 151 | bestCircleRadius = 0 152 | bestCirclePercentage = 0 153 | 154 | minCirclePercentage = 0.6 # at least 60% of a circle must be present 155 | 156 | maxNrOfIterations = len(edgePositions) # TODO: adjust this parameter or include some real ransac criteria with inlier/outlier percentages to decide when to stop 157 | 158 | for its in xrange(maxNrOfIterations): 159 | # RANSAC: randomly choose 3 point and create a circle: 160 | 161 | # TODO: choose randomly but more intelligent, 162 | # so that it is more likely to choose three points of a circle. 163 | # For example if there are many small circles, it is unlikely to randomly choose 3 points of the same circle. 164 | idx1 = np.random.randint(len(edgePositions)) 165 | idx2 = np.random.randint(len(edgePositions)) 166 | idx3 = np.random.randint(len(edgePositions)) 167 | 168 | # we need 3 different samples: 169 | if idx1 == idx2 or idx1 == idx3 or idx3 == idx2: 170 | continue 171 | 172 | # create circle from 3 points: 173 | center, radius = getCircle(edgePositions[idx1], edgePositions[idx2], edgePositions[idx3]) 174 | if not center or radius > maxRadius: 175 | continue 176 | 177 | # inlier set unused at the moment but could be used to approximate a (more robust) circle from alle inlier 178 | # verify or falsify the circle by inlier counting: 179 | cPerc, inlierSet = verifyCircle(dt, center, radius) 180 | 181 | if cPerc >= bestCirclePercentage and radius >= minRadius: 182 | bestCirclePercentage = cPerc 183 | bestCircleRadius = radius 184 | bestCircleCenter = center 185 | 186 | # draw if good circle was found 187 | if bestCirclePercentage >= minCirclePercentage and bestCircleRadius >= minRadius: 188 | cv2.circle(mask, (int(round(bestCircleCenter[0])), int(round(bestCircleCenter[1]))), int(round(bestCircleRadius)), 255, -1) 189 | # mask found circle 190 | cv2.circle(edges, (int(round(bestCircleCenter[0])), int(round(bestCircleCenter[1]))), int(round(bestCircleRadius)), (0, 0, 0), 3) 191 | return mask 192 | 193 | 194 | def weightMap(hue, intensity, edges, roi): 195 | ellipseMask, contourMask = findEllipses(edges) 196 | circlesMask = findCircles(hue, intensity) 197 | ransacMask = findRANSACCircles(edges) 198 | 199 | # create a map by combining different masks 200 | # circle/ellipse detection masks have a higher weight then roi mask and ransac mask 201 | combinedMask = np.zeros(edges.shape, dtype=np.float32) 202 | combinedMask += ellipseMask 203 | combinedMask += contourMask 204 | combinedMask += circlesMask 205 | combinedMask += ransacMask.astype(np.float32) / 2.0 206 | combinedMask += roi.astype(np.float32) / 4.0 207 | 208 | nonZeroMask = combinedMask != 0 209 | 210 | # rescale mask to 0 .. 1 range 211 | # (weight is either in 0.25 .. 1 range or 0) 212 | combinedMask[nonZeroMask] /= 255 * 3.75 213 | combinedMask[nonZeroMask] *= 0.6 214 | combinedMask[nonZeroMask] += 0.4 215 | return combinedMask 216 | 217 | 218 | def roiMask(image, boundaries): 219 | scale = max([1.0, np.average(np.array(image.shape)[0:2] / 400.0)]) 220 | shape = (int(round(image.shape[1] / scale)), int(round(image.shape[0] / scale))) 221 | 222 | small_color = cv2.resize(image, shape, interpolation=cv2.INTER_LINEAR) 223 | 224 | # reduce details and remove noise for better edge detection 225 | small_color = cv2.bilateralFilter(small_color, 8, 64, 64) 226 | small_color = cv2.pyrMeanShiftFiltering(small_color, 8, 64, maxLevel=1) 227 | small = cv2.cvtColor(small_color, cv2.COLOR_BGR2HSV) 228 | 229 | hue = small[::, ::, 0] 230 | intensity = cv2.cvtColor(small_color, cv2.COLOR_BGR2GRAY) 231 | 232 | edges = extractEdges(hue, intensity) 233 | roi = roiFromEdges(edges) 234 | weight_map = weightMap(hue, intensity, edges, roi) 235 | 236 | _, final_mask = cv2.threshold(roi, 5, 255, cv2.THRESH_BINARY) 237 | small = cv2.bitwise_and(small, small, mask=final_mask) 238 | 239 | kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4)) 240 | 241 | for (lower, upper) in boundaries: 242 | lower = np.array([lower, 80, 50], dtype="uint8") 243 | upper = np.array([upper, 255, 255], dtype="uint8") 244 | 245 | # find the colors within the specified boundaries and apply 246 | # the mask 247 | mask = cv2.inRange(small, lower, upper) 248 | mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel, iterations=3) 249 | mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel, iterations=1) 250 | final_mask = cv2.bitwise_and(final_mask, mask) 251 | 252 | # blur the mask for better contour extraction 253 | final_mask = cv2.GaussianBlur(final_mask, (5, 5), 0) 254 | return (final_mask, weight_map, scale) 255 | 256 | 257 | def extractRoi(image, winSize, stepSize): 258 | # hue boundaries 259 | colors = [ 260 | (15, 30) # orange-yellow 261 | ] 262 | 263 | mask, weight_map, mask_scale = roiMask(image, colors) 264 | contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) 265 | yield weight_map, mask_scale 266 | 267 | for resized in pyramid(image, winSize): 268 | scale = image.shape[0] / resized.shape[0] 269 | for x, y, w, h in boundingRects(mask_scale, contours): 270 | x /= scale 271 | y /= scale 272 | w /= scale 273 | h /= scale 274 | center = (min(x + w / 2, resized.shape[1]), min(y + h / 2, resized.shape[0])) 275 | if w > winSize[0] or h > winSize[1]: 276 | for x, y, window in sliding_window(resized, (int(x), int(y), int(w), int(h)), stepSize, winSize): 277 | yield ((x, y, winSize[0], winSize[1]), scale, window) 278 | else: 279 | x = max(0, int(center[0] - winSize[0] / 2)) 280 | y = max(0, int(center[1] - winSize[1] / 2)) 281 | window = resized[y:y + winSize[1], x:x + winSize[0]] 282 | yield ((x, y, winSize[0], winSize[1]), scale, window) 283 | -------------------------------------------------------------------------------- /TemplateMatcher.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import numpy as np 3 | from multiprocessing.pool import ThreadPool 4 | from sklearn.base import BaseEstimator, TransformerMixin 5 | 6 | FLANN_INDEX_KDTREE = 1 # bug: flann enums are missing 7 | 8 | 9 | class Templates(): 10 | def __init__(self, features): 11 | self.features = features 12 | 13 | 14 | class TemplateMatch(TransformerMixin, BaseEstimator): 15 | def __init__(self, templates, ratio=0.75): 16 | self.templates = templates 17 | self.ratio = ratio 18 | 19 | flann_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5) 20 | self.matcher = cv2.FlannBasedMatcher(flann_params, {}) 21 | self.pool = ThreadPool(processes=cv2.getNumberOfCPUs()) 22 | 23 | def __getstate__(self): 24 | return dict(templates=self.templates, ratio=self.ratio) 25 | 26 | def __setstate__(self, state): 27 | self.templates = state['templates'] 28 | self.ratio = state['ratio'] 29 | 30 | flann_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5) 31 | self.matcher = cv2.FlannBasedMatcher(flann_params, {}) 32 | self.pool = ThreadPool(processes=1) # cv2.getNumberOfCPUs()) 33 | 34 | def feature(self, match): 35 | return match.distance 36 | 37 | def process(self, features): 38 | count = len(features) 39 | if not count: 40 | return [0, 0, 0, 0, 0, 0] 41 | 42 | split = [0.08, 0.12096, 0.16192, 0.20288, 0.24384, 0.2848, 1] 43 | return np.histogram(features, bins=split)[0] 44 | 45 | def match(self, sample): 46 | feature = [] 47 | for template in self.templates.features: 48 | # we have to have at least 2 descriptors for 2 nearest neighbour search 49 | if len(sample) < 2: 50 | feature.extend([0, 0, 0, 0, 0, 0]) 51 | else: 52 | if template.dtype != sample.dtype or template.shape[1] != sample.shape[1]: 53 | print '!!!' 54 | print template.shape, template.dtype, sample.shape, sample.dtype 55 | raw_matches = self.matcher.knnMatch(template, trainDescriptors=sample, k=2) 56 | matches = [self.feature(m[0]) 57 | for m in raw_matches 58 | if len(m) == 2 and m[0].distance < m[1].distance * self.ratio] 59 | feature.extend(self.process(matches)) 60 | return np.array(feature) 61 | 62 | def fit(self, X, y=None, **fit_params): 63 | return self 64 | 65 | def transform(self, X, **transform_params): 66 | l = dict(n=0) 67 | 68 | def f(image): 69 | l['n'] += 1 70 | if (100 * l['n'] / len(X)) % 5 == 0: 71 | print '\r%d%% - %d/%d' % (100 * l['n'] / len(X), l['n'], len(X)), 72 | return self.match(image) 73 | 74 | ires = self.pool.imap(f, X) 75 | return np.array(list(ires)) 76 | -------------------------------------------------------------------------------- /avery_8254.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/AVGInnovationLabs/DoNotSnap/e679947fd3bdfec7c9c786ee285e9ac64d8d1a18/avery_8254.pdf -------------------------------------------------------------------------------- /classifier.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/AVGInnovationLabs/DoNotSnap/e679947fd3bdfec7c9c786ee285e9ac64d8d1a18/classifier.pdf -------------------------------------------------------------------------------- /classify.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import sys 3 | import pickle 4 | import numpy as np 5 | 6 | from AffineInvariantFeatures import AffineInvariant 7 | from RegionOfInterest import extractRoi 8 | from TemplateMatcher import TemplateMatch, Templates # imported to allow deserialization 9 | from util import non_max_suppression_fast 10 | 11 | from PIL import Image 12 | from matplotlib import pyplot as plt 13 | 14 | 15 | def classify(model, features, coords, weight_map, mask_scale): 16 | rects = [] 17 | 18 | if len(features): 19 | y_pred = model.predict_proba(features) 20 | success = 0 21 | 22 | for (_, result), ((x, y, w, h), scale) in zip(y_pred, coords): 23 | mask_window = weight_map[y / mask_scale:(y + h) / mask_scale, x / mask_scale:(x + w) / mask_scale] 24 | weight = mask_window.max() 25 | if result * weight > 0.5: 26 | success += 1 27 | rects.append([x * scale, y * scale, (x + w) * scale, (y + h) * scale]) 28 | 29 | print 'windows identified as positive: %d/%d(%0.2f%%)' % (success, len(y_pred), 100.0 * success / len(y_pred)) 30 | else: 31 | print 'windows identified as positive: 0/%d(0%%)' % len(features) 32 | 33 | return np.array(rects) 34 | 35 | 36 | def main(image_file): 37 | image = Image.open(image_file) 38 | if image is None: 39 | print 'Could not load image "%s"' % sys.argv[1] 40 | return 41 | 42 | image = np.array(image.convert('RGB'), dtype=np.uint8) 43 | image = image[:, :, ::-1].copy() 44 | 45 | winSize = (200, 200) 46 | stepSize = 32 47 | 48 | roi = extractRoi(image, winSize, stepSize) 49 | weight_map, mask_scale = next(roi) 50 | 51 | samples = [(rect, scale, cv2.cvtColor(window, cv2.COLOR_BGR2GRAY)) 52 | for rect, scale, window in roi] 53 | 54 | X_test = [window for rect, scale, window in samples] 55 | coords = [(rect, scale) for rect, scale, window in samples] 56 | 57 | extractor = cv2.FeatureDetector_create('SURF') 58 | detector = cv2.DescriptorExtractor_create('SURF') 59 | 60 | affine = AffineInvariant(extractor, detector) 61 | 62 | saved = pickle.load(open('classifier.pkl', 'rb')) 63 | 64 | feature_transform = saved['pipe'] 65 | model = saved['model'] 66 | 67 | print 'Extracting Affine transform invariant features' 68 | affine_invariant_features = affine.transform(X_test) 69 | print 'Matching features with template' 70 | features = feature_transform.transform(affine_invariant_features) 71 | 72 | rects = classify(model, features, coords, weight_map, mask_scale) 73 | for (left, top, right, bottom) in non_max_suppression_fast(rects, 0.4): 74 | cv2.rectangle(image, (left, top), (right, bottom), (0, 0, 0), 10) 75 | cv2.rectangle(image, (left, top), (right, bottom), (32, 32, 255), 5) 76 | 77 | plt.imshow(cv2.cvtColor(image, cv2.COLOR_BGR2RGB)) 78 | plt.show() 79 | 80 | if __name__ == '__main__': 81 | image_file = sys.argv[1] if len(sys.argv) >= 2 else 'sample.jpg' 82 | main(image_file) 83 | -------------------------------------------------------------------------------- /dns_badge.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/AVGInnovationLabs/DoNotSnap/e679947fd3bdfec7c9c786ee285e9ac64d8d1a18/dns_badge.png -------------------------------------------------------------------------------- /dns_logo.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/AVGInnovationLabs/DoNotSnap/e679947fd3bdfec7c9c786ee285e9ac64d8d1a18/dns_logo.png -------------------------------------------------------------------------------- /sample.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/AVGInnovationLabs/DoNotSnap/e679947fd3bdfec7c9c786ee285e9ac64d8d1a18/sample.jpg -------------------------------------------------------------------------------- /setup.py: -------------------------------------------------------------------------------- 1 | from distutils.core import setup 2 | import py2exe 3 | import matplotlib 4 | 5 | setup(console = ['classify.py'], 6 | data_files = matplotlib.get_py2exe_datafiles(), 7 | options = dict( 8 | py2exe = dict( 9 | compressed = True, 10 | optimize = 2, 11 | dll_excludes = ['MSVCP90.dll'], 12 | packages = ['FileDialog'], 13 | includes = [ 14 | 'scipy', 15 | 'scipy.integrate', 16 | 'scipy.special.*', 17 | 'scipy.linalg.*', 18 | 'scipy.sparse.csgraph._validation', 19 | 'matplotlib', 20 | 'matplotlib.backends.backend_tkagg', 21 | 'matplotlib.pyplot', 22 | 'sklearn', 23 | 'sklearn.pipeline', 24 | 'sklearn.tree', 25 | 'sklearn.tree._utils', 26 | 'sklearn.utils.lgamma' 27 | ], 28 | excludes = [ 29 | '_gtkagg', 30 | '_tkagg', 31 | '_agg2', 32 | '_cairo', 33 | '_cocoaagg', 34 | '_fltkagg', 35 | '_gtk', 36 | '_gtkcairo', 37 | 'tcl' 38 | ] 39 | ) 40 | )) 41 | -------------------------------------------------------------------------------- /template.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/AVGInnovationLabs/DoNotSnap/e679947fd3bdfec7c9c786ee285e9ac64d8d1a18/template.png -------------------------------------------------------------------------------- /train.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import sys 3 | import pickle 4 | import numpy as np 5 | import matplotlib.pyplot as plt 6 | 7 | from AffineInvariantFeatures import AffineInvariant 8 | from TemplateMatcher import TemplateMatch, Templates 9 | 10 | from PIL import Image 11 | from itertools import izip_longest 12 | 13 | from sklearn.cross_validation import train_test_split 14 | from sklearn.grid_search import GridSearchCV 15 | from sklearn.metrics import classification_report, roc_curve, auc, confusion_matrix, accuracy_score 16 | from sklearn.pipeline import Pipeline 17 | from sklearn.tree import export_graphviz, DecisionTreeClassifier 18 | 19 | 20 | def line_count(filename): 21 | with open(filename) as data: 22 | return sum(1 for line in data) 23 | 24 | 25 | def read_image(filename): 26 | return np.array(Image.open(filename.strip('\n')).convert('L'), np.uint8) 27 | 28 | 29 | def read_file(filename, limit=0): 30 | n = 0 31 | lines = line_count(filename) 32 | 33 | with open(filename) as data: 34 | while True: 35 | line = next(data, None) 36 | if not line or (limit and n >= limit): 37 | break 38 | 39 | n += 1 40 | print '\r%s %d/%d' % (filename, n, limit or lines), 41 | try: 42 | yield read_image(line) 43 | except: 44 | continue 45 | 46 | 47 | def get_templates(): 48 | return np.array(list(read_file('templates.txt'))) 49 | 50 | 51 | def get_images(limit=0): 52 | positive = read_file('positive.txt', limit / 2 if limit else 0) 53 | negative = read_file('negative.txt', limit / 2 if limit else 0) 54 | 55 | for p, n in izip_longest(positive, negative): 56 | if p is not None: 57 | yield (1, p) 58 | if n is not None: 59 | yield (0, n) 60 | 61 | 62 | def get_dataset(limit): 63 | return map(np.asarray, zip(*get_images(limit))) 64 | 65 | 66 | def plot_roc(fpr, tpr, roc_auc): 67 | # Plot all ROC curves 68 | plt.figure() 69 | 70 | plt.plot(fpr, tpr, label='ROC curve (area = %0.2f)' % roc_auc) 71 | 72 | plt.plot([0, 1], [0, 1], 'k--') 73 | plt.xlim([0.0, 1.0]) 74 | plt.ylim([0.0, 1.05]) 75 | plt.xlabel('False Positive Rate') 76 | plt.ylabel('True Positive Rate') 77 | plt.title('Affine Invariant SURF + Decision Tree Classifier') 78 | plt.legend(loc='lower right') 79 | plt.show() 80 | 81 | 82 | def plot_importance(feature_count, importances, indices): 83 | plt.figure() 84 | plt.title('Feature importances') 85 | plt.bar(range(feature_count), importances[indices], color='r', align='center') 86 | plt.xticks(range(feature_count), indices) 87 | plt.xlim([-1, feature_count]) 88 | plt.show() 89 | 90 | 91 | def main(name, dataset_size): 92 | templates = get_templates() 93 | 94 | print 'templates: %d' % len(templates) 95 | 96 | labels, samples = get_dataset(dataset_size) 97 | 98 | print 'samples: %d' % len(samples) 99 | 100 | extractor = cv2.FeatureDetector_create('SURF') 101 | detector = cv2.DescriptorExtractor_create('SURF') 102 | 103 | print 'applying affine invariant transform' 104 | 105 | affine = AffineInvariant(extractor, detector) 106 | templates = affine.transform(templates) 107 | samples = affine.transform(samples) 108 | 109 | model = Pipeline([ 110 | ('match', TemplateMatch(Templates(templates))), # XXX: hack to bypass cloning error 111 | # ('reduce_dim', PCA(n_components = 12 * 6)) 112 | ]) 113 | 114 | samples = model.fit_transform(samples) 115 | 116 | rng = np.random.RandomState() 117 | 118 | X_train, X_test, y_train, y_test = train_test_split(samples, labels, test_size=0.5, random_state=rng) 119 | print 'train: %d, test: %d' % (len(X_train), len(X_test)) 120 | 121 | params = dict( 122 | min_samples_split = [5, 6, 7, 8, 9, 10], 123 | min_samples_leaf = [3, 4, 5, 6, 7], 124 | max_leaf_nodes = [10, 9, 8, 7, 6], 125 | class_weight = [{1: w} for w in [10, 8, 4, 2, 1]] 126 | ) 127 | 128 | tree = DecisionTreeClassifier(max_depth=4, random_state=rng) 129 | 130 | cvmodel = GridSearchCV(tree, params, cv=10, n_jobs=cv2.getNumberOfCPUs()) 131 | cvmodel.fit(X_train, y_train) 132 | 133 | print 'grid scores' 134 | for params, mean_score, scores in cvmodel.grid_scores_: 135 | print '%0.3f (+/-%0.03f) for %r' % (mean_score, scores.std() * 2, params) 136 | print 'best parameters' 137 | print cvmodel.best_params_ 138 | 139 | importances = cvmodel.best_estimator_.feature_importances_ 140 | indices = np.argsort(importances)[::-1] 141 | 142 | plot_importance(6, importances, indices) 143 | 144 | y_pred = cvmodel.predict(X_test) 145 | accuracy = accuracy_score(y_test, y_pred) 146 | 147 | print 'accuracy: %f' % accuracy 148 | print classification_report(y_test, y_pred) 149 | print confusion_matrix(y_test, y_pred) 150 | 151 | y_score = cvmodel.predict_proba(X_test)[:, 1] 152 | fpr, tpr, _ = roc_curve(y_test, y_score) 153 | roc_auc = auc(fpr, tpr) 154 | 155 | plot_roc(fpr, tpr, roc_auc) 156 | 157 | export_graphviz(cvmodel.best_estimator_, out_file=name + '.dot', class_names=['background', 'badge'], filled=True, rounded=True, special_characters=True) 158 | pickle.dump(dict(params=params, pipe=model, model=cvmodel.best_estimator_), open(name + '.pkl', 'wb')) 159 | 160 | 161 | if __name__ == '__main__': 162 | name = sys.argv[1] if len(sys.argv) >= 2 else 'classifier' 163 | dataset_size = int(sys.argv[2]) if len(sys.argv) >= 3 else 0 164 | 165 | main(name, dataset_size) 166 | -------------------------------------------------------------------------------- /util.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import numpy as np 3 | 4 | 5 | def sliding_window(image, start, stepSize, windowSize): 6 | width = max(1, start[2] - windowSize[0]) 7 | height = max(1, start[3] - windowSize[1]) 8 | # slide a window across the image 9 | for y in xrange(start[1], start[1] + height, stepSize): 10 | for x in xrange(start[0], start[0] + width, stepSize): 11 | h = min(windowSize[1], start[3]) 12 | w = min(windowSize[0], start[2]) 13 | center = (min(x + w / 2, image.shape[1]), min(y + h / 2, image.shape[0])) 14 | x = max(0, int(center[0] - windowSize[0] / 2)) 15 | y = max(0, int(center[1] - windowSize[1] / 2)) 16 | # yield the current window 17 | yield (x, y, image[y:y + windowSize[1], x:x + windowSize[0]]) 18 | 19 | 20 | def pyramid(image, minSize): 21 | yield image 22 | 23 | if image.shape[0] < minSize[0] and image.shape[1] < minSize[1]: 24 | # image too small - upscaling until we hit window level 25 | image = cv2.pyrUp(image) 26 | 27 | while (image.shape[0] <= minSize[0] or image.shape[1] <= minSize[1]): 28 | yield image 29 | image = cv2.pyrUp(image) 30 | else: 31 | # image too big - downscaling until we hit window level 32 | image = cv2.pyrDown(image) 33 | 34 | while (image.shape[0] >= minSize[0] or image.shape[1] >= minSize[1]): 35 | yield image 36 | image = cv2.pyrDown(image) 37 | 38 | 39 | # Malisiewicz et al. 40 | def non_max_suppression_fast(boxes, overlapThresh): 41 | # if there are no boxes, return an empty list 42 | if len(boxes) == 0: 43 | return [] 44 | 45 | # if the bounding boxes integers, convert them to floats -- 46 | # this is important since we'll be doing a bunch of divisions 47 | if boxes.dtype.kind == "i": 48 | boxes = boxes.astype("float") 49 | 50 | # initialize the list of picked indexes 51 | pick = [] 52 | 53 | # grab the coordinates of the bounding boxes 54 | x1 = boxes[:, 0] 55 | y1 = boxes[:, 1] 56 | x2 = boxes[:, 2] 57 | y2 = boxes[:, 3] 58 | 59 | # compute the area of the bounding boxes and sort the bounding 60 | # boxes by the bottom-right y-coordinate of the bounding box 61 | area = (x2 - x1 + 1) * (y2 - y1 + 1) 62 | idxs = np.argsort(y2) 63 | 64 | # keep looping while some indexes still remain in the indexes 65 | # list 66 | while len(idxs) > 0: 67 | # grab the last index in the indexes list and add the 68 | # index value to the list of picked indexes 69 | last = len(idxs) - 1 70 | i = idxs[last] 71 | pick.append(i) 72 | 73 | # find the largest (x, y) coordinates for the start of 74 | # the bounding box and the smallest (x, y) coordinates 75 | # for the end of the bounding box 76 | xx1 = np.maximum(x1[i], x1[idxs[:last]]) 77 | yy1 = np.maximum(y1[i], y1[idxs[:last]]) 78 | xx2 = np.minimum(x2[i], x2[idxs[:last]]) 79 | yy2 = np.minimum(y2[i], y2[idxs[:last]]) 80 | 81 | # compute the width and height of the bounding box 82 | w = np.maximum(0, xx2 - xx1 + 1) 83 | h = np.maximum(0, yy2 - yy1 + 1) 84 | 85 | # compute the ratio of overlap 86 | overlap = (w * h) / area[idxs[:last]] 87 | 88 | # delete all indexes from the index list that have 89 | idxs = np.delete(idxs, 90 | np.concatenate(([last], np.where(overlap > overlapThresh)[0]))) 91 | 92 | # return only the bounding boxes that were picked using the 93 | # integer data type 94 | return boxes[pick].astype("int") 95 | 96 | 97 | # Felzenszwalb et al. 98 | def non_max_suppression_slow(boxes, overlapThresh): 99 | # if there are no boxes, return an empty list 100 | if len(boxes) == 0: 101 | return [] 102 | 103 | # initialize the list of picked indexes 104 | pick = [] 105 | 106 | # grab the coordinates of the bounding boxes 107 | x1 = boxes[:, 0] 108 | y1 = boxes[:, 1] 109 | x2 = boxes[:, 2] 110 | y2 = boxes[:, 3] 111 | 112 | # compute the area of the bounding boxes and sort the bounding 113 | # boxes by the bottom-right y-coordinate of the bounding box 114 | area = (x2 - x1 + 1) * (y2 - y1 + 1) 115 | idxs = np.argsort(y2) 116 | 117 | # keep looping while some indexes still remain in the indexes 118 | # list 119 | while len(idxs) > 0: 120 | # grab the last index in the indexes list, add the index 121 | # value to the list of picked indexes, then initialize 122 | # the suppression list (i.e. indexes that will be deleted) 123 | # using the last index 124 | last = len(idxs) - 1 125 | i = idxs[last] 126 | pick.append(i) 127 | suppress = [last] 128 | 129 | # loop over all indexes in the indexes list 130 | for pos in xrange(0, last): 131 | # grab the current index 132 | j = idxs[pos] 133 | 134 | # find the largest (x, y) coordinates for the start of 135 | # the bounding box and the smallest (x, y) coordinates 136 | # for the end of the bounding box 137 | xx1 = max(x1[i], x1[j]) 138 | yy1 = max(y1[i], y1[j]) 139 | xx2 = min(x2[i], x2[j]) 140 | yy2 = min(y2[i], y2[j]) 141 | 142 | # compute the width and height of the bounding box 143 | w = max(0, xx2 - xx1 + 1) 144 | h = max(0, yy2 - yy1 + 1) 145 | 146 | # compute the ratio of overlap between the computed 147 | # bounding box and the bounding box in the area list 148 | overlap = float(w * h) / area[j] 149 | 150 | # if there is sufficient overlap, suppress the 151 | # current bounding box 152 | if overlap > overlapThresh: 153 | suppress.append(pos) 154 | 155 | # delete all indexes from the index list that are in the 156 | # suppression list 157 | idxs = np.delete(idxs, suppress) 158 | 159 | # return only the bounding boxes that were picked 160 | return boxes[pick] 161 | --------------------------------------------------------------------------------