用keras做简单的人脸识别

原文网址：https://blog.csdn.net/eagelangel/article/details/50759993

原文存在几个小问题：本文在此基础上做了一下修改，方便大家调试程序

首先基于theano和TensorFlow的输入格式不一样，同时由于keras版本不同，在进行模型compile和fit时，所需要的格式也不一样；本文keras的backend为：TensorFlow，需要注意的部分，已用红色进行标注

code:

from __future__ import print_function  
import numpy  
numpy.random.seed(1337)  # for reproducibility  
  
from PIL import Image  
  
from keras.models import Sequential  
from keras.layers.core import Dense, Dropout, Activation, Flatten  
from keras.layers.convolutional import Convolution2D, MaxPooling2D  
from keras.optimizers import SGD  
from keras.utils import np_utils  
  
# There are 40 different classes  
nb_classes = 40  
nb_epoch = 40  
batch_size = 40  
  
# input image dimensions  
img_rows, img_cols = 57, 47  
# number of convolutional filters to use  
nb_filters1, nb_filters2 = 5, 10  
# size of pooling area for max pooling  
nb_pool = 2  
# convolution kernel size  
nb_conv = 3  
  
def load_data(dataset_path):  
    img = Image.open(dataset_path)  
    img_ndarray = numpy.asarray(img, dtype='float64')/256  
    #400pictures,size:57*47=2679  
    faces=numpy.empty((400,2679))   
    for row in range(20):  
       for column in range(20):  
        faces[row*20+column]=numpy.ndarray.flatten(img_ndarray [row*57:(row+1)*57,column*47:(column+1)*47])  
  
    label=numpy.empty(400)  
    for i in range(40):  
      label[i*10:i*10+10]=i  
    label=label.astype(numpy.int)  
  
    #train:320,valid:40,test:40  
    train_data=numpy.empty((320,2679))  
    train_label=numpy.empty(320)  
    valid_data=numpy.empty((40,2679))  
    valid_label=numpy.empty(40)  
    test_data=numpy.empty((40,2679))  
    test_label=numpy.empty(40)  
  
    for i in range(40):  
      train_data[i*8:i*8+8]=faces[i*10:i*10+8]  
      train_label[i*8:i*8+8]=label[i*10:i*10+8]  
      valid_data[i]=faces[i*10+8]  
      valid_label[i]=label[i*10+8]  
      test_data[i]=faces[i*10+9]  
      test_label[i]=label[i*10+9]  
       
    rval = [(train_data, train_label), (valid_data, valid_label),  
            (test_data, test_label)]  
    return rval  
  
def Net_model(lr=0.005,decay=1e-6,momentum=0.9):  
    model = Sequential()  
    model.add(Convolution2D(nb_filters1, nb_conv, nb_conv,  
                            border_mode='valid',  
                            input_shape=(img_rows, img_cols,  1)))  
    model.add(Activation('tanh'))  
    model.add(MaxPooling2D(pool_size=(nb_pool, nb_pool)))  
  
    model.add(Convolution2D(nb_filters2, nb_conv, nb_conv))  
    model.add(Activation('tanh'))  
    model.add(MaxPooling2D(pool_size=(nb_pool, nb_pool)))  
    model.add(Dropout(0.25))  
  
    model.add(Flatten())  
    model.add(Dense(1000)) #Full connection  
    model.add(Activation('tanh'))  
    model.add(Dropout(0.5))  
    model.add(Dense(nb_classes))  
    model.add(Activation('softmax'))  
  
    sgd = SGD(lr=lr, decay=decay, momentum=momentum, nesterov=True)  
    model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])  
      
    return model  
  
def train_model(model,X_train,Y_train,X_val,Y_val):  
    model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,  
            verbose=1, validation_data=(X_val, Y_val)) 
    model.save_weights('model_weights.h5',overwrite=True)  
    return model  
  
def test_model(model,X,Y):  
    model.load_weights('model_weights.h5')  
    score = model.evaluate(X, Y, verbose=0)
    print('Test score:', score[0])  
    print('Test accuracy:', score[1])  
    return score  
  
if __name__ == '__main__':  
    # the data, shuffled and split between tran and test sets  
    (X_train, y_train), (X_val, y_val),(X_test, y_test) = load_data('olivettifaces.gif')  
  
    X_train = X_train.reshape(X_train.shape[0], img_rows, img_cols, 1)  
    X_val = X_val.reshape(X_val.shape[0], img_rows, img_cols, 1)  
    X_test = X_test.reshape(X_test.shape[0], img_rows, img_cols, 1)  
    print('X_train shape:', X_train.shape)  
    print(X_train.shape[0], 'train samples')  
    print(X_val.shape[0], 'validate samples')  
    print(X_test.shape[0], 'test samples')  
  
    # convert class vectors to binary class matrices  
    Y_train = np_utils.to_categorical(y_train, nb_classes)  
    Y_val = np_utils.to_categorical(y_val, nb_classes)  
    Y_test = np_utils.to_categorical(y_test, nb_classes)  
  
    model=Net_model()  
    train_model(model,X_train,Y_train,X_val,Y_val)   
    score=test_model(model,X_test,Y_test) 
     
  
    model.load_weights('model_weights.h5')  
    classes=model.predict_classes(X_test,verbose=0)  
    test_accuracy = numpy.mean(numpy.equal(y_test,classes))  

    print("accuarcy:",test_accuracy) 

最终的程序运行结果示意图：