pca人脸识别

2018-05-07

目标

对于人脸识别问题，通过pca降维后进行分类。

数据来源

att_faces 数据集：

40个人的黑白人脸，每个人有10张图片，每个人随机抽取5张图片，共200张图片经过pca变化之后作为训练集

用1NN，和kernal=linear的svm两种方法进行分类，识别率差不多在95%左右

数据下载地址 http://www.cl.cam.ac.uk/Research/DTG/attarchive:pub/data/att_faces.tar.Z

实现大概过程

通过opencv加载图片的像素点数据,图片所在的文件目录作为图片的类标
pca变化(计算均值->中心化->归一化)得到训练集
1-NN 预测测试集类标， svm预测测试集类标
判断识别正确率

代码实现

#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time    : 2018/5/6 下午9:50
# @Author  : Junnor.G
# @Site    : NJU
# @File    : pca_face_recognition.py
# @Software: PyCharm
# -*- coding: utf-8 -*-
import numpy as np
import os, glob, random, cv2
from sklearn.svm import SVC
import matplotlib.pyplot as plt
def pca(data,k):
    print('shape: {}'.format(data.shape))
    data = np.float32(np.mat(data))
    rows,cols = data.shape                              #取大小
    data_mean = np.mean(data, 0)                         #求均值
    Z = data - np.tile(data_mean, (rows,1))
    D,V = np.linalg.eig(Z*Z.T)                         #特征值与特征向量
    V1 = V[:, :k]                                       #取前k个特征向量
    V1 = Z.T*V1
    for i in range(k):                                 #特征向量归一化
        V1[:,i] /= np.linalg.norm(V1[:,i])
    return np.array(Z*V1),data_mean,V1
def loadImageSet(folder=u'orl_faces/', sampleCount=5): #加载图像集，随机选择sampleCount张图片用于训练
    trainData = []; testData = []; yTrain=[]; yTest = [];
    for k in range(40):
        folder2 = os.path.join(folder, 's%d' % (k+1))
        # print('debug:{}'.format(folder2))
        # print('debug:{}'.format(glob.glob(os.path.join(folder2, '*.pgm'))))
        data = [cv2.imread(d, 0) for d in glob.glob(os.path.join(folder2, '*.pgm'))]
        sample = random.sample(range(10), sampleCount)
        trainData.extend([data[i].ravel() for i in range(10) if i in sample])
        testData.extend([data[i].ravel() for i in range(10) if i not in sample])
        yTest.extend([k]* (10-sampleCount))
        yTrain.extend([k]* sampleCount)
    return np.array(trainData),  np.array(yTrain), np.array(testData), np.array(yTest)
def plot(x):
    fig = plt.figure()
    # gs = gridspec.GridSpec(4, 4)
    # gs.update(wspace=0.05, hspace=0.05)
    ax = plt.subplot(111)
    plt.axis('off')
    ax.set_xticklabels([])
    ax.set_yticklabels([])
    ax.set_aspect('equal')
    plt.imshow(x.reshape(14, 14), cmap='Greys_r')
    return fig
def main():
    xTrain_, yTrain, xTest_, yTest = loadImageSet()
    num_train, num_test = xTrain_.shape[0], xTest_.shape[0]
    xTrain,data_mean,V = pca(xTrain_, 14*14)
    xTest = np.array((xTest_-np.tile(data_mean,(num_test,1))) * V)  #得到测试脸在特征向量下的数据
    fig = plot(xTrain[0])
    plt.savefig('{}.png'.format('generate'), bbox_inches='tight')
    print('XTRAIN: {}'.format(xTrain[0]))
    print('YTrain: {}'.format(yTrain))
    yPredict =[yTrain[np.sum((xTrain-np.tile(d,(num_train,1)))**2, 1).argmin()] for d in xTest]
    # print('predict: {}'.format(yPredict))
    # print('really:  {}'.format(np.array(yTest)))
    print (u'欧式距离法识别率: %.2f%%'% ((yPredict == np.array(yTest)).mean()*100))
    svm = SVC(kernel='linear')                    #支持向量机方法
    svm.fit(np.array(xTrain), np.float32(yTrain))
    yPredict = svm.predict(np.float32(xTest))
    # print('predict: {}'.format(yPredict))
    # print('really:  {}'.format(np.array(yTest)))
    #yPredict = svm.predict_all(xTest.astype(np.float64))
    print (u'支持向量机识别率: %.2f%%' % ((yPredict == np.array(yTest)).mean()*100))
if __name__ =='__main__':
    main()