資源簡(jiǎn)介
基于HOG特征提取的圖像分類(lèi)器,HOG的核心思想是所檢測(cè)的局部物體外形能夠被光強(qiáng)梯度或邊緣方向的分布所描述。通過(guò)將整幅圖像分割成小的連接區(qū)域稱(chēng)為cells,每個(gè)cell生成一個(gè)方向梯度直方圖或者cell中pixel的邊緣方向,這些直方圖的組合可表示出所檢測(cè)目標(biāo)的目標(biāo))描述子。為改善準(zhǔn)確率,局部直方圖可以通過(guò)計(jì)算圖像中一個(gè)較大區(qū)域稱(chēng)為block的光強(qiáng)作為measure被對(duì)比標(biāo)準(zhǔn)化,然后用這個(gè)measure歸一化這個(gè)block中的所有cells.這個(gè)歸一化過(guò)程完成了更好的照射/陰影不變性。

代碼片段和文件信息
function?blockfeat?=?BinHOGFeature(?b_magb_orientcell_sizenblock...??
????bin_num?weight_vote)??
%?計(jì)算1個(gè)block的hog??
%?weight_vote:?是否進(jìn)行高斯加權(quán)投票??
???
%?block的HOG直方圖??
blockfeat=zeros(bin_num*nblock^21);??
???
%?高斯權(quán)重??
gaussian_weight=fspecial(‘gaussian‘cell_size*nblock0.5*cell_size*nblock);??
???
%?分割block??
for?n=1:nblock??
????for?m=1:nblock??
????????%?cell的左上角坐標(biāo)??
????????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-1x_off:x_off+cell_size-1);??
????????c_orient=b_orient(y_off:y_off+cell_size-1x_off:x_off+cell_size-1);??
???
????????%?cell的hog直方圖??
????????c_feat=zeros(bin_num1);??
????????for?i=1:bin_num??
????????????%?是否進(jìn)行高斯加權(quán)?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直方圖中??
????????count=(n-1)*nblock+m;??
????????blockfeat((count-1)*bin_num+1:count*bin_num1)=c_feat;??
????end??
end??
???
%?歸一化?L2-norm??
sump=sum(blockfeat.^2);??
blockfeat?=?blockfeat./sqrt(sump+eps^2);??
?屬性????????????大小?????日期????時(shí)間???名稱(chēng)
-----------?---------??----------?-----??----
?????文件???????2243??2016-12-09?15:23??ImgHOGFeature.m
?????文件??????15335??2016-09-23?16:15??3.jpg
?????文件???????1307??2016-12-09?15:04??BinHOGFeature.m
-----------?---------??----------?-----??----
????????????????18885????????????????????3
評(píng)論
共有 條評(píng)論