【Python深度学习】Tensorflow+CNN进行人脸识别实战（附源码和数据集）

需要源码和数据集请点赞关注收藏后评论区留言私信~~~

下面利用tensorflow平台进行人脸识别实战，使用的是Olivetti Faces人脸图像 部分数据集展示如下

 程序训练过程如下

 接下来训练CNN模型 可以看到训练进度和损失值变化

接下来展示人脸识别结果

程序会根据一张图片自动去图片集中寻找相似的人脸 如上图所示

部分代码如下 需要全部源码和数据集请点赞关注收藏后评论区留言私信~~~

1. from os import listdir
2. import numpy as np
3. from PIL import Image
4. import cv2
5. from tensorflow.keras.models import Sequential, load_model
6. from tensorflow.keras.layers import Dense, Activation, Convolution2D, MaxPooling2D, Flatten
7. from sklearn.model_selection import train_test_split
8. from tensorflow.python.keras.utils import np_utils
10. # 读取人脸图片数据
11. def img2vector(fileNamestr):
12. # 创建向量
13. returnVect = np.zeros((57,47))
14. image = Image.open(fileNamestr).convert('L')
15. img = np.asarray(image).reshape(57,47)
16. return img
18. # 制作人脸数据集
19. def GetDataset(imgDataDir):
20. print('| Step1 |: Get dataset...')
21. imgDataDir='faces_4/'
22. FileDir = listdir(imgDataDir)
24. m = len(FileDir)
25. imgarray=[]
26. hwLabels=[]
27. hwdata=[]
29. # 逐个读取图片文件
30. for i in range(m):
31. # 提取子目录
32. className=i
33. subdirName='faces_4/'+str(FileDir[i])+'/'
34. fileNames = listdir(subdirName)
35. lenFiles=len(fileNames)
36. # 提取文件名
37. for j in range(lenFiles):
38. fileNamestr = subdirName+fileNames[j]
39. hwLabels.append(className)
40. imgarray=img2vector(fileNamestr)
41. hwdata.append(imgarray)
43. hwdata = np.array(hwdata)
44. return hwdata,hwLabels,6
46. # CNN模型类
47. class MyCNN(object):
48. FILE_PATH = "face_recognition.h5" # 模型存储/读取目录
49. picHeight = 57 # 模型的人脸图片长47，宽57
50. picWidth = 47
52. def __init__(self):
53. self.model = None
55. # 获取训练数据集
56. def read_trainData(self, dataset):
57. self.dataset = dataset
59. # 建立Sequential模型，并赋予参数
60. def build_model(self):
61. print('| Step2 |: Init CNN model...')
62. self.model = Sequential()
63. print('self.dataset.X_train.shape[1:]',self.dataset.X_train.shape[1:])
64. self.model.add( Convolution2D( filters=32,
65. kernel_size=(5, 5),
66. padding='same',
67. #dim_ordering='th',
68. input_shape=self.dataset.X_train.shape[1:]))
70. self.model.add(Activation('relu'))
71. self.model.add( MaxPooling2D(pool_size=(2, 2),
72. strides=(2, 2),
73. padding='same' ) )
74. self.model.add(Convolution2D(filters=64,
75. kernel_size=(5, 5),
76. padding='same') )
77. self.model.add(Activation('relu'))
78. self.model.add(MaxPooling2D(pool_size=(2, 2),
79. strides=(2, 2),
80. padding='same') )
81. self.model.add(Flatten())
82. self.model.add(Dense(512))
83. self.model.add(Activation('relu'))
85. self.model.add(Dense(self.dataset.num_classes))
86. self.model.add(Activation('softmax'))
87. self.model.summary()
89. # 模型训练
90. def train_model(self):
91. print('| Step3 |: Train CNN model...')
92. self.model.compile( optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
93. # epochs：训练代次、batch_size：每次训练样本数
94. self.model.fit(self.dataset.X_train, self.dataset.Y_train, epochs=10, batch_size=20)
96. def evaluate_model(self):
97. loss, accuracy = self.model.evaluate(self.dataset.X_test, self.dataset.Y_test)
98. print('| Step4 |: Evaluate performance...')
99. print('===================================')
100. print('Loss Value is :', loss)
101. print('Accuracy Value is :', accuracy)
103. def save(self, file_path=FILE_PATH):
104. print('| Step5 |: Save model...')
105. self.model.save(file_path)
106. print('Model ',file_path,'is succeesfuly saved.')
108. # 建立一个用于存储和格式化读取训练数据的类
109. class DataSet(object):
110. def __init__(self, path):
111. self.num_classes = None
112. self.X_train = None
113. self.X_test = None
114. self.Y_train = None
115. self.Y_test = None
116. self.picWidth = 47
117. self.picHeight = 57
118. self.makeDataSet(path) # 在这个类初始化的过程中读取path下的训练数据
120. def makeDataSet(self, path):
121. # 根据指定路径读取出图片、标签和类别数
122. imgs, labels, clasNum = GetDataset(path)
124. # 将数据集打乱随机分组
125. X_train, X_test, y_train, y_test = train_test_split(imgs, labels, test_size=0.2,random_state=1)
127. # 重新格式化和标准化
128. X_train = X_train.reshape(X_train.shape[0], 1, self.picHeight, self.picWidth) / 255.0
129. X_test = X_test.reshape(X_test.shape[0], 1, self.picHeight, self.picWidth) / 255.0
131. X_train = X_train.astype('float32')
132. X_test = X_test.astype('float32')
134. # 将labels转成 binary class matrices
135. Y_train = np_utils.to_categorical(y_train, num_classes=clasNum)
136. Y_test = np_utils.to_categorical(y_test, num_classes=clasNum)
138. # 将格式化后的数据赋值给类的属性上
139. self.X_train = X_train
140. self.X_test = X_test
141. self.Y_train = Y_train
142. self.Y_test = Y_test
143. self.num_classes = clasNum
144. # 人脸图片目录
145. dataset = DataSet('faces_4/')
146. model = MyCNN()
147. model.read_trainData(dataset)
148. model.build_model()
149. model.train_model()
150. model.evaluate_model()
151. model.save()

 创作不易 觉得有帮助请点赞关注收藏~~~