wavelet feature 128D 小波特征

function f=WTF(I);
% compute the 128-D wavelet transform feature of image I
% Input:
% I: the input image
% Output:
% f:128-D feature
%%warning:normalization must be performed after all the features are extracted from the dataset  

%I=imread('D:/2.jpg');
if(ndims(I)==3)I=rgb2gray(I);end

wname='db8';
[Lo_D,Hi_D,Lo_R,Hi_R] = wfilters(wname);

scale=3;

%%PWT feature,24-D array (4+4+4)*2
cA=I;
f=[];
for i=1:scale
	[c,s]= wavedec2(cA,1,Lo_D,Hi_D);
	msize=s(1,1)*s(1,2);
	cA=c(1,1:msize);
	%imshow(uint8(cA));
	cA=reshape(cA,s(1,1),s(1,2));
	%cH cV cD
	cH=c(1,msize+(1:msize));
	cV=c(1,2*msize+(1:msize));
	cD=c(1,3*msize+(1:msize));
	cH=reshape(cH,s(2,1),s(2,2));
	cV=reshape(cV,s(2,1),s(2,2));
	cD=reshape(cD,s(2,1),s(2,2));
	
	%figure,colormap gray
	%subplot(221),imagesc(cA);
	%subplot(222),imagesc(cH);
	%subplot(223),imagesc(cV);
	%subplot(224),imagesc(cD);
	%X   = waverec2(c,s,'db8');
	%f=[mean(mean(cA) std2(cA)];
	f=[f mean2((abs(cA))) std2(abs(cA)) mean2((abs(cH))) std2(abs(cH)) mean2((abs(cV))) std2(abs(cV)) mean2((abs(cD))) std2(abs(cD))];
	%f=[f mean2(((cA))) std2((cA)) mean2(((cH))) std2((cH)) mean2(((cV))) std2((cV)) mean2(((cD))) std2((cD))];
end
f1=f;

f=[];

%%TWT feature,104-D 4*(1+3+9)*2 array
set=[];
candidate={};
cA=I;
set{1}=cA;%%store the frequency (LL,LH,HL) to be decomposed
candidate{1}=cA;%%

for j=1:scale
	count=0;
	set=candidate;
	for i=1:length(set)
			cA=set{i};
		%for i=1:3
			[c,s]= wavedec2(cA,1,Lo_D,Hi_D);
			msize=s(1,1)*s(1,2);
			cA=c(1,1:msize);
			%imshow(uint8(cA));
			cA=reshape(cA,s(1,1),s(1,2));
			%cH cV cD

			cH=c(1,msize+(1:msize));
			cV=c(1,2*msize+(1:msize));
			cD=c(1,3*msize+(1:msize));%%HH level 
			cH=reshape(cH,s(2,1),s(2,2));
			cV=reshape(cV,s(2,1),s(2,2));
			cD=reshape(cD,s(2,1),s(2,2));%%HH
			count=count+1;
			candidate{count}=cA;
			count=count+1;
			candidate{count}=cH;
			count=count+1;
			candidate{count}=cV;
			%figure,colormap gray
			%subplot(221),imagesc(cA);
			%subplot(222),imagesc(cH);
			%subplot(223),imagesc(cV);
			%subplot(224),imagesc(cD);
			%%%%%X   = waverec2(c,s,'db8');
			%f=[mean(mean(cA) std2(cA)];
			f=[f mean2((abs(cA))) std2(abs(cA)) mean2((abs(cH))) std2(abs(cH)) mean2((abs(cV))) std2(abs(cV)) mean2((abs(cD))) std2(abs(cD))];
		%end
	end
	count
end
f2=f;

f=[f1 f2];

参考资料：

【1】NUS-WIDE: A Real-World Web Image Database from，National University of Singapore

【2】Texture Features for Browsing and Retrieval of Image Data

【3】Texture Analysis and Classification with Tree-structured Wavelet Transform