Source Code for Module pyvision.face.SVMFaceRec

 1  import pyvision as pv 
 2  from pyvision.vector.SVM import SVM,TYPE_C_SVC 
 3  #from pyvision.vector.VectorClassifier import NORM_NONE 
 4  #from pyvision.vector.PCA import PCA 
 5  from scipy import ndimage 
 6  #import random 
 7   
 8   
 9 -class SVMFaceRec: 
10       
11 -    def __init__(self): 
12          self.tile_size = (128,160) 
13          self.leye = pv.Point(26.0,40.0) 
14          self.reye = pv.Point(102.0,40.0) 
15          self.norm_sigma = 8.0 
16           
17          self.svm = None 
18           
19          self.n_faces = 0 
20          self.n_labels = 0 
21           
22          self.training_data = {} 
23           
24               
25 -    def preprocess(self,im,leye,reye,ilog=None): 
26          im = pv.Image(im.asPIL()) 
27          affine = pv.AffineFromPoints(leye,reye,self.leye,self.reye,self.tile_size) 
28          tile = affine.transformImage(im) 
29           
30          mat = tile.asMatrix2D() 
31           
32          # High pass filter the image 
33          mat = mat - ndimage.gaussian_filter(mat,self.norm_sigma) 
34           
35          # Value normalize the image. 
36          mat = mat - mat.mean() 
37          mat = mat / mat.std() 
38                   
39          tile = pv.Image(mat) 
40           
41          return tile 
42           
43 -    def addTraining(self,im,leye,reye,sub_id,ilog=None): 
44          self.svm = None 
45           
46          tile = self.preprocess(im,leye,reye,ilog) 
47          if not self.training_data.has_key(sub_id): 
48              self.training_data[sub_id] = [] 
49           
50          self.training_data[sub_id].append(tile) 
51   
52          self.n_labels = len(self.training_data) 
53          self.n_faces += 1 
54           
55           
56 -    def train(self, C = None, Gamma = None, ilog=None, callback=None): 
57          # Create the SVM 
58          self.svm = SVM(svm_type=TYPE_C_SVC) 
59           
60          # Add training data 
61          for sub_id, tiles in self.training_data.iteritems(): 
62              for tile in tiles: 
63                  self.svm.addTraining(sub_id,tile) 
64                   
65          # Train the SVM 
66          if C != None and Gamma != None: 
67              self.svm.train(C_range = C, G_range = Gamma, verbose=True,callback=callback) 
68          else: 
69              #Automatic 
70              self.svm.train(verbose=True,callback=callback) 
71               
72   
73           
74 -    def predict(self,im,leye,reye): 
75          assert self.svm != None 
76           
77          tile = self.preprocess(im,leye,reye) 
78          return self.svm.predict(tile) 
79   
80 -    def score(self,im,leye,reye): 
81          tile = self.preprocess(im,leye,reye) 
82          return self.svm.predictSVMValues(tile) 
83       
84 -    def reset(self): 
85          self.svm = None 
86           
87          self.n_faces = 0 
88          self.n_labels = 0 
89           
90          self.training_data = {} 
91           
92 -    def isTrained(self): 
93          return self.svm != None 
94