python机器学习库keras——CNN卷积神经网络人脸识别

全栈工程师开发手册 &＃xff08;作者&＃xff1a;栾鹏&＃xff09;
python教程全解

github地址&＃xff1a;https://github.com/626626cdllp/kears/tree/master/Face_Recognition

图片来源

图片中共40个人&＃xff0c;每人10张图片&＃xff0c;每张图片高57&＃xff0c;宽47。共400张图片。

读取图片的py文件

import numpy
import pandas
from PIL import Image
from keras import backend as K
from keras.utils import np_utils"""
加载图像数据的函数,dataset_path即图像olivettifaces的路径
加载olivettifaces后&＃xff0c;划分为train_data,valid_data,test_data三个数据集
函数返回train_data,valid_data,test_data以及对应的label
"""# 400个样本&＃xff0c;40个人&＃xff0c;每人10张样本图。每张样本图高57*宽47&＃xff0c;需要2679个像素点。每个像素点做了归一化处理
def load_data(dataset_path):img &＃61; Image.open(dataset_path)img_ndarray &＃61; numpy.asarray(img, dtype&＃61;&＃39;float64&＃39;) / 256print(img_ndarray.shape)faces &＃61; numpy.empty((400,57,47))for row in range(20):for column in range(20):faces[row * 20 &＃43; column] &＃61; img_ndarray[row * 57:(row &＃43; 1) * 57, column * 47:(column &＃43; 1) * 47]# 设置400个样本图的标签label &＃61; numpy.empty(400)for i in range(40):label[i * 10:i * 10 &＃43; 10] &＃61; ilabel &＃61; label.astype(numpy.int)label &＃61; np_utils.to_categorical(label, 40) # 将40分类类标号转化为one-hot编码# 分成训练集、验证集、测试集&＃xff0c;大小如下train_data &＃61; numpy.empty((320, 57,47)) # 320个训练样本train_label &＃61; numpy.empty((320,40)) # 320个训练样本&＃xff0c;每个样本40个输出概率valid_data &＃61; numpy.empty((40, 57,47)) # 40个验证样本valid_label &＃61; numpy.empty((40,40)) # 40个验证样本&＃xff0c;每个样本40个输出概率test_data &＃61; numpy.empty((40, 57,47)) # 40个测试样本test_label &＃61; numpy.empty((40,40)) # 40个测试样本&＃xff0c;每个样本40个输出概率for i in range(40):train_data[i * 8:i * 8 &＃43; 8] &＃61; faces[i * 10:i * 10 &＃43; 8]train_label[i * 8:i * 8 &＃43; 8] &＃61; label[i * 10:i * 10 &＃43; 8]valid_data[i] &＃61; faces[i * 10 &＃43; 8]valid_label[i] &＃61; label[i * 10 &＃43; 8]test_data[i] &＃61; faces[i * 10 &＃43; 9]test_label[i] &＃61; label[i * 10 &＃43; 9]return [(train_data, train_label), (valid_data, valid_label),(test_data, test_label)]if __name__ &＃61;&＃61; &＃39;__main__&＃39;:[(train_data, train_label), (valid_data, valid_label), (test_data, test_label)] &＃61; load_data(&＃39;olivettifaces.gif&＃39;)oneimg &＃61; train_data[0]*256print(oneimg)im &＃61; Image.fromarray(oneimg)im.show()


CNN人脸识别代码

import numpy as np
np.random.seed(1337) # for reproducibility
from keras.models import Sequential
from keras.layers import Dense, Activation, Flatten
from keras.layers import Conv2D, MaxPooling2D,AveragePooling2D
from PIL import Image
import FaceData
# 全局变量
batch_size &＃61; 128 # 批处理样本数量
nb_classes &＃61; 40 # 分类数目
epochs &＃61; 600 # 迭代次数
img_rows, img_cols &＃61; 57, 47 # 输入图片样本的宽高
nb_filters &＃61; 32 # 卷积核的个数
pool_size &＃61; (2, 2) # 池化层的大小
kernel_size &＃61; (5, 5) # 卷积核的大小
input_shape &＃61; (img_rows, img_cols,1) # 输入图片的维度[(X_train, Y_train), (X_valid, Y_valid),(X_test, Y_test)] &＃61;FaceData.load_data(&＃39;olivettifaces.gif&＃39;)X_train&＃61;X_train[:,:,:,np.newaxis] # 添加一个维度&＃xff0c;代表图片通道。这样数据集共4个维度&＃xff0c;样本个数、宽度、高度、通道数
X_valid&＃61;X_valid[:,:,:,np.newaxis] # 添加一个维度&＃xff0c;代表图片通道。这样数据集共4个维度&＃xff0c;样本个数、宽度、高度、通道数
X_test&＃61;X_test[:,:,:,np.newaxis] # 添加一个维度&＃xff0c;代表图片通道。这样数据集共4个维度&＃xff0c;样本个数、宽度、高度、通道数print(&＃39;样本数据集的维度&＃xff1a;&＃39;, X_train.shape,Y_train.shape)
print(&＃39;测试数据集的维度&＃xff1a;&＃39;, X_test.shape,Y_test.shape)# 构建模型
model &＃61; Sequential()
model.add(Conv2D(6,kernel_size,input_shape&＃61;input_shape,strides&＃61;1)) # 卷积层1
model.add(AveragePooling2D(pool_size&＃61;pool_size,strides&＃61;2)) # 池化层
model.add(Conv2D(12,kernel_size,strides&＃61;1)) # 卷积层2
model.add(AveragePooling2D(pool_size&＃61;pool_size,strides&＃61;2)) # 池化层
model.add(Flatten()) # 拉成一维数据
model.add(Dense(nb_classes)) # 全连接层2
model.add(Activation(&＃39;sigmoid&＃39;)) # sigmoid评分# 编译模型
model.compile(loss&＃61;&＃39;categorical_crossentropy&＃39;,optimizer&＃61;&＃39;adadelta&＃39;,metrics&＃61;[&＃39;accuracy&＃39;])
# 训练模型
model.fit(X_train, Y_train, batch_size&＃61;batch_size, epochs&＃61;epochs,verbose&＃61;1, validation_data&＃61;(X_test, Y_test))
# 评估模型
score &＃61; model.evaluate(X_test, Y_test, verbose&＃61;0)
print(&＃39;Test score:&＃39;, score[0])
print(&＃39;Test accuracy:&＃39;, score[1])y_pred &＃61; model.predict(X_test)
y_pred &＃61; y_pred.argmax(axis&＃61;1) # 获取概率最大的分类&＃xff0c;获取每行最大值所在的列
for i in range(len(y_pred)):oneimg &＃61; X_test[i,:,:,0]*256im &＃61; Image.fromarray(oneimg)im.show()print(&＃39;第%d个人识别为第%d个人&＃39;%(i,y_pred[i]))

600次迭代&＃xff0c;正确率90%&＃xff0c;当然只用了10个样本进行测试&＃xff0c;所以准确率不是特别准确。