matlab人脸识别基于pca和bp神经网络的代码实现

以下是基于PCA和BP神经网络的MATLAB人脸识别代码实现，其中包括数据预处理、特征提取、模型训练和测试等步骤。

1. 数据预处理

首先需要准备训练数据和测试数据。数据集可以使用公开的人脸数据库，如Yale人脸数据库、ORL人脸数据库等。这里以Yale人脸数据库为例，该数据库包含15个人的165张灰度图像，每个人有11张不同表情的图像。代码如下：

clear all; clc;
% 读取数据
dataDir = 'yalefaces';
imgList = dir(fullfile(dataDir,'*.*'));
imgNum = length(imgList);
imgSize = [243, 320]; % 图像大小
imgData = zeros(imgSize(1)*imgSize(2), imgNum);
for i = 1:imgNum
    img = imread(fullfile(dataDir, imgList(i).name));
    img = imresize(img, imgSize);
    imgData(:,i) = img(:);
end

% 数据归一化
imgData = double(imgData);
imgData = imgData - mean(imgData, 2); % 减去均值
imgData = imgData ./ std(imgData, 0, 2); % 归一化

2. 特征提取

接下来，使用PCA方法对数据进行降维，提取出最重要的特征。代码如下：

% PCA降维
[U,S,V] = svd(imgData, 'econ');
eigVals = diag(S).^2;
energy = cumsum(eigVals) / sum(eigVals);
thres = find(energy >= 0.99, 1);
U = U(:,1:thres);
feaData = U.' * imgData;

3. 模型训练

使用BP神经网络对特征进行分类。首先，将数据集分为训练集和测试集，代码如下：

% 数据集分割
trainNum = 10; % 每个人的训练样本数
testNum = 11 - trainNum; % 每个人的测试样本数
trainData = zeros(size(feaData,1), trainNum*15);
trainLabel = zeros(15, trainNum*15);
testData = zeros(size(feaData,1), testNum*15);
testLabel = zeros(15, testNum*15);
for i = 1:15
    idx = (i-1)*11+1:i*11;
    trainData(:,(i-1)*trainNum+1:i*trainNum) = feaData(:,idx(1:trainNum));
    trainLabel(i,(i-1)*trainNum+1:i*trainNum) = 1;
    testData(:,(i-1)*testNum+1:i*testNum) = feaData(:,idx(trainNum+1:end));
    testLabel(i,(i-1)*testNum+1:i*testNum) = 1;
end

然后，搭建BP神经网络模型并进行训练。代码如下：

% BP神经网络训练
net = feedforwardnet([20,10]);
net.trainFcn = 'trainlm';
net.trainParam.show = 50;
net.trainParam.epochs = 1000;
net.trainParam.goal = 1e-5;
net.trainParam.lr = 0.01;
[net, tr] = train(net, trainData, trainLabel);

4. 模型测试

最后，使用测试数据对模型进行测试，并计算识别准确率。代码如下：

% BP神经网络测试
testOutput = net(testData);
[~, testPred] = max(testOutput);
[~, testTarget] = max(testLabel);
accuracy = sum(testPred == testTarget) / length(testTarget);
fprintf('Accuracy: %.2f%%\n', accuracy*100);

完整代码如下：

clear all; clc;
% 读取数据
dataDir = 'yalefaces';
imgList = dir(fullfile(dataDir,'*.*'));
imgNum = length(imgList);
imgSize = [243, 320]; % 图像大小
imgData = zeros(imgSize(1)*imgSize(2), imgNum);
for i = 1:imgNum
    img = imread(fullfile(dataDir, imgList(i).name));
    img = imresize(img, imgSize);
    imgData(:,i) = img(:);
end

% 数据归一化
imgData = double(imgData);
imgData = imgData - mean(imgData, 2); % 减去均值
imgData = imgData ./ std(imgData, 0, 2); % 归一化

% PCA降维
[U,S,V] = svd(imgData, 'econ');
eigVals = diag(S).^2;
energy = cumsum(eigVals) / sum(eigVals);
thres = find(energy >= 0.99, 1);
U = U(:,1:thres);
feaData = U.' * imgData;

% 数据集分割
trainNum = 10; % 每个人的训练样本数
testNum = 11 - trainNum; % 每个人的测试样本数
trainData = zeros(size(feaData,1), trainNum*15);
trainLabel = zeros(15, trainNum*15);
testData = zeros(size(feaData,1), testNum*15);
testLabel = zeros(15, testNum*15);
for i = 1:15
    idx = (i-1)*11+1:i*11;
    trainData(:,(i-1)*trainNum+1:i*trainNum) = feaData(:,idx(1:trainNum));
    trainLabel(i,(i-1)*trainNum+1:i*trainNum) = 1;
    testData(:,(i-1)*testNum+1:i*testNum) = feaData(:,idx(trainNum+1:end));
    testLabel(i,(i-1)*testNum+1:i*testNum) = 1;
end

% BP神经网络训练
net = feedforwardnet([20,10]);
net.trainFcn = 'trainlm';
net.trainParam.show = 50;
net.trainParam.epochs = 1000;
net.trainParam.goal = 1e-5;
net.trainParam.lr = 0.01;
[net, tr] = train(net, trainData, trainLabel);

% BP神经网络测试
testOutput = net(testData);
[~, testPred] = max(testOutput);
[~, testTarget] = max(testLabel);
accuracy = sum(testPred == testTarget) / length(testTarget);
fprintf('Accuracy: %.2f%%\n', accuracy*100);