hog特征提取matlab代码

以下是使用MATLAB提取HOG特征的示例代码。假设有一张图像img，图像大小为[height,width]，并且希望将图像划分为n个单元格。

% 定义参数
cellSize = 8; % 单元格大小
numBins = 9; % 直方图方向数
height = size(img,1);
width = size(img,2);

% 计算梯度
I = rgb2gray(img);
Ix = conv2(I,[-1 0 1],'same');
Iy = conv2(I,[-1;0;1],'same');
grad = sqrt(Ix.^2 + Iy.^2);
angle = atan2(Iy,Ix);

% 将角度转化为0-180度
angle(angle<0) = angle(angle<0) + pi;

% 将图像划分为单元格
numHorzCells = floor(width/cellSize);
numVertCells = floor(height/cellSize);
histograms = zeros(numVertCells,numHorzCells,numBins);

% 计算每个单元格的直方图
for i=1:numVertCells
    for j=1:numHorzCells
        rowStart = (i-1)*cellSize+1;
        rowEnd = i*cellSize;
        colStart = (j-1)*cellSize+1;
        colEnd = j*cellSize;
        histogram = zeros(numBins,1);
        for m=rowStart:rowEnd
            for n=colStart:colEnd
                bin = floor(numBins*angle(m,n)/pi)+1;
                histogram(bin) = histogram(bin) + grad(m,n);
            end
        end
        histograms(i,j,:) = histogram;
    end
end

% 将每个单元格的直方图拼接为特征向量
features = reshape(histograms,[],numBins);
features = bsxfun(@rdivide,features,sum(features,2)+eps);
features = sqrt(features);
features = bsxfun(@rdivide,features,norm(features)+eps);

其中，bsxfun函数是一种广播函数，用于处理两个具有不同大小的矩阵之间的操作。