HOG特征提取

实现过程：

先举个例子：对于64*128的图像而言，每16*16的像素组成一个cell，每2*2个cell组成一个块，因为每个cell有9个特征，所以每个块内有4*9=36个，以8个像素为步长，那么，水平方向将有7个扫描窗口，垂直方向将有15个扫描窗口。也就是说，64*128的图片，总特征共有36*7*15=3780个特征。

步骤：

参考网友MATLAB程序

function HOGFeature = ImgHOGFeature( imgPathName, cell_size, nblock,...,    overlap, angle, bin_num)% 计算输入图像的特征描述子% imgPathName：图片路径% cell_size: cell的长宽% nblock: block的width、height包括的cell的个数% overlap: block重叠的比例% angle: 180\360% bin_num: 方向bin的数目 if nargin<2    % default parameter    cell_size=8;    nblock=2;    overlap=0.5;    angle=180;    bin_num=9;elseif nargin<6    error('Input parameters are not enough.');end Img = imread('F:\壁纸\honey.jpg');imshow(Img);if size(Img,3) == 3    % 简化计算，直接转换成灰度图像    G = rgb2gray(Img);else    G = Img;end figure,imshow(G);[height, width] = size(G); % 计算x、y方向的梯度hx = [-1,0,1];hy = -hx';grad_x = imfilter(double(G),hx);grad_y = imfilter(double(G),hy);figure,imshow(grad_x);figure,imshow(grad_y);% 计算梯度的模长grad_mag=sqrt(grad_x.^2+grad_y.^2); % 计算梯度的方向index= grad_x==0;grad_x(index)=1e-5;YX=grad_y./grad_x;if angle==180    grad_angle= ((atan(YX)+(pi/2))*180)./pi; elseif angle==360    grad_angle= ((atan2(grad_y,grad_x)+pi).*180)./pi;end % orient binbin_angle=angle/bin_num;grad_orient=ceil(grad_angle./bin_angle); % 计算block的个数block_size=cell_size*nblock;skip_step=block_size*overlap;x_step_num=floor((width-block_size)/skip_step+1);   %x方向窗口个数y_step_num=floor((height-block_size)/skip_step+1); % 初始化hog特征描述子feat_dim=bin_num*nblock^2;HOGFeature=zeros(feat_dim,x_step_num*y_step_num); for k=1:y_step_num    for j=1:x_step_num        % block的左上角坐标        x_off = (j-1)*skip_step+1;        y_off = (k-1)*skip_step+1;         % 取得block的梯度大小和方向        b_mag=grad_mag(y_off:y_off+block_size-1,x_off:x_off+block_size-1);        b_orient=grad_orient(y_off:y_off+block_size-1,x_off:x_off+block_size-1);         % 当前block的hog直方图        currFeat = BinHOGFeature(b_mag, b_orient, cell_size,nblock, bin_num, false);        HOGFeature(:, (k-1)*x_step_num+j) = currFeat;     endend end

其中，BinHOGFeature为

function blockfeat = BinHOGFeature( b_mag,b_orient,cell_size,nblock,...    bin_num, weight_vote)% 计算1个block的hog% weight_vote: 是否进行高斯加权投票 % block的HOG直方图blockfeat=zeros(bin_num*nblock^2,1); % 高斯权重gaussian_weight=fspecial('gaussian',cell_size*nblock,0.5*cell_size*nblock); % 分割blockfor n=1:nblock    for m=1:nblock        % cell的左上角坐标        x_off = (m-1)*cell_size+1;        y_off = (n-1)*cell_size+1;         % cell的梯度大小和方向        c_mag=b_mag(y_off:y_off+cell_size-1,x_off:x_off+cell_size-1);        c_orient=b_orient(y_off:y_off+cell_size-1,x_off:x_off+cell_size-1);         % cell的hog直方图        c_feat=zeros(bin_num,1);        for i=1:bin_num            % 是否进行高斯加权 投票            if weight_vote==false                c_feat(i)=sum(c_mag(c_orient==i));            else                c_feat(i)=sum(c_mag(c_orient==i).*gaussian_weight(c_orient==i));            end        end         % 合并到block的HOG直方图中        count=(n-1)*nblock+m;        blockfeat((count-1)*bin_num+1:count*bin_num,1)=c_feat;    endend % 归一化 L2-normsump=sum(blockfeat.^2);blockfeat = blockfeat./sqrt(sump+eps^2);

小白一枚，为以后方便再看

参考链接，感谢

http://blog.csdn.net/zouxy09/article/details/7929348
http://blog.csdn.net/woxincd/article/details/18351807