├── README.md
└── brisque_features.py


/README.md:
--------------------------------------------------------------------------------
1 | # IQA-python
2 | Python code to compute features of classic Image Quality Assessment models
3 | 


--------------------------------------------------------------------------------
/brisque_features.py:
--------------------------------------------------------------------------------
 1 | # Python code for BRISQUE model
 2 | # Original paper title: No-Reference Image Quality Assessment in the Spatial Domain
 3 | # Link: http://ieeexplore.ieee.org/document/6272356/
 4 | import cv2
 5 | import numpy as np
 6 | from scipy import ndimage
 7 | import math
 8 |     
 9 | def get_gaussian_filter():
10 |     [m,n] = [(ss - 1.0) / 2.0 for ss in (shape,shape)]
11 |     [y,x] = np.ogrid[-m:m+1,-n:n+1]
12 |     window = np.exp( -(x*x + y*y) / (2.0*sigma*sigma) )
13 |     window[window < np.finfo(window.dtype).eps*window.max() ] = 0
14 |     sum_window = window.sum()
15 |     if sum_window != 0:
16 |         window = np.divide(window, sum_window)
17 |     return window
18 | 
19 | def lmom(X):
20 |     (rows, cols)  = X.shape
21 |     if cols == 1:
22 |         X = X.reshape(1,rows)
23 |     n = rows
24 |     X.sort()    
25 |     b = np.zeros(3)    
26 |     b0 = X.mean()    
27 |     for r in range(1,4):        
28 |         Num = np.prod(np.tile(np.arange(r+1,n+1), (r,1))-np.tile(np.arange(1,r+1).reshape(r,1),(1,n-r)),0)        
29 |         Num = Num.astype(np.float)                
30 |         Den = np.prod(np.tile(n, (1, r)) - np.arange(1,r+1), 1)        
31 |         b[r-1] = 1.0/n * sum(Num/Den * X[0,r:])
32 |     L = np.zeros(4)
33 |     L[0] = b0
34 |     L[1] = 2*b[0] - b0
35 |     L[2] = 6*b[1] - 6*b[0] + b0
36 |     L[3] = 20*b[2] - 30*b[1] + 12*b[0] - b0
37 |     return L
38 | 
39 | def compute_features(im):
40 |     im = im.astype(np.float)
41 |     window = get_gaussian_filter()
42 |     scalenum = 2
43 |     feat = []
44 |     for itr_scale in range(scalenum):
45 |         mu = cv2.filter2D(im, cv2.CV_64F, window, borderType=cv2.BORDER_CONSTANT)
46 |         mu_sq = mu * mu
47 |         sigma = np.sqrt(abs(cv2.filter2D(im*im, cv2.CV_64F, window, borderType=cv2.BORDER_CONSTANT) - mu_sq))        
48 |         structdis = (im-mu)/(sigma+1)
49 |         structdis_col_vector = np.reshape(structdis.transpose(), (structdis.size,1))
50 |         L = lmom(structdis.reshape(structdis.size,1))
51 |         feat = np.append(feat,[L[1], L[3]])
52 |         shifts = [[0,1], [1,0], [1,1], [-1,1]]
53 |         for itr_shift in shifts:
54 |             shifted_structdis = np.roll(structdis, itr_shift[0], axis=0)
55 |             shifted_structdis = np.roll(shifted_structdis, itr_shift[1], axis=1)
56 | 
57 |             shifted_structdis_col_vector = np.reshape(shifted_structdis.T, (shifted_structdis.size,1))
58 |             pair = structdis_col_vector * shifted_structdis_col_vector
59 |             L = lmom(pair.reshape(pair.size,1))
60 |             feat = np.append(feat, L)
61 |         im = cv2.resize(im, (0,0), fx=0.5, fy=0.5, interpolation=cv2.INTER_CUBIC)
62 |     return feat
63 | 
64 | 
65 | im = ndimage.imread('example.bmp', flatten=True)
66 | feat = compute_features(im)
67 | print feat
68 | 


--------------------------------------------------------------------------------