adding the GLCM features (divide the image into 16-by-16 blocks, and for each block get the energy and contrast features)

dear brothers and sisters

i am new student in matlab i have a task that dataset (contains 2 categories, each has 50 image),

i -calculate the same features as above for each image

2- divide the data into 25 images train and 25 images test for each category

3- calculate the % correct in retrieving the nearest image for each query

4- adding the GLCM features (divide the image into 16-by-16 blocks, and for each block get the energy and contrast features)

5- Also, try to use the PCA

----------------------------------------------

i finsh points (1,2,3)

how can i applay GLCM and PCA

ANY ONE HELP

i will uploud my code which solve points (1,2,3) which we make it in two file

first file ---------------->

%This code reads all the COREL images and extracts feature vector for each image

%The features used are:

%(a)

clear;

clearvars;

%First we load the total image that contains images from all categories

%Then we create the color map from this image

for jndex=1:2 %over the 2 categories

if(jndex==1)

wwww = 'NPatsmall';

elseif(jndex==2)

wwww = 'TPatsmall';

end

www= wwww ;

K=40; %the DCT block width

KK=16; %the histohram length of R,G,B

KKK = 256; %this is the size of the Global color MAP

%here we load an image which is a combination of many colors

ZZZ=imread('TOTAL.jpg'); %this is the combined image

[IIND MAP] = rgb2ind(ZZZ,256); %this is the global color table MAP

ind=0;

path=strcat('Patches\', www ,'\');

outputfile2=strcat(www);

outputfileName2=strcat(path,outputfile2);

outputfileName3=strcat(outputfile2);

for iiii=1:50 %loop over the 100 images in each category

iii = (jndex-1)*50 + iiii -1;

fileName1=strcat(num2str(iii,'%2d'),'.tif');

outputfile1=strcat(num2str(iii,'%2d'),'.feats');

fileName1=strcat(path,fileName1);

outputfileName=strcat(path,outputfile1);

%reading the image:

X= imread(fileName1);

Xg=rgb2gray(X);

%This is the DCT2 for the GRAY and we take the upper left DCT2 coefficients

%with a square of size K-by-K

Xg=im2double(Xg);

Jg = dct2(Xg);

jg=im2col(Jg,[K K],'distinct'); %if K=12 this gives a vector length 144

bg = jg(:,1);

%For K=40, this will give 1600 features

%Here we get each color channel alone

y1=X(:,:,1); %Red

y2=X(:,:,2); %Green

y3=X(:,:,3); %Blue

Y1=im2double(y1);

Y2=im2double(y2);

Y3=im2double(y3);

%Now for the global color histograms for each separate color channel

h1 = imhist(Y1,KK); %red histogram

h2 = imhist(Y2,KK); %green histogram

h3 = imhist(Y3,KK); %blue histogram

%each histo has only KK bins (chosen to be 16 here)

h=[h1; h2; h3]; %here we concatenate the 3 histos from the 3 colors

%this makes 3*16 = 48 features in this case

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%Now we have the color histogram for the combined colors using the color

%MAP. This color map was obtained from a collection of one image from each

%category. The MAP has

Xind = rgb2ind(X,MAP); %this transforms the image to an indexed image using the total map

UU = imhist(Xind,MAP); %this is the KKK vector with the combined colors histogram

%here KKK=256 giving 256 features

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

D= [bg; h; UU]; %the concatenated super vector

%So, total number of features is 144+48+256 = 448 features.

L = length(D);

ind=ind+1;

Data(ind,1:L)=D(1:L);

end

save(outputfileName3,'Data','-ASCII'); %this saves one feature file for each category

%in the same folder that this file is working from.

end

seconf file -------------------->

%in this code we do locally adative random projections followed by a

%bagging trees selection and retrieval

clear;

clearvars;

doPCA=1; %this is to do PCA on the RAW features to produce reduced dimension features

NC=21; %# of PCA coeffs.

Normalize=0; %normalize each feature before the PCA: here we only divide over the max(abs)

ZNormalize=0; %if 1 then Z-normalize each feature after PCA

UseBias= 0; %to add a bias term after the PCA

HII = 50; %the maximum number of images to recall

%The COREL data with some not very good features (indexed color histogram, DCT,

%color histograms)

load NPatsmall

load TPatsmall

sss=size(NPatsmall); %they all have same size

s11 = sss(1); %this is the number of examples in this category (100,true50)

s22 = sss(2); %this is the number of features extracted from the original images(3)

AAAA = [NPatsmall; TPatsmall]; %This array contains the whole data. Size= (1000,true100) by #features

mam=max(abs(AAAA)); %the absolute max for each feature

%this is to make feature ranges close to each others

s2=s22;

if(Normalize==1)

for i=1:100

for j=1:s2

AA(i,j) = AAAA(i,j)/mam(j);

end

end

else

AA = AAAA;

end

%Now we divide the data into training and testing (50,true 25)% each

cn1=0;

cn2=0;

for i=1:2:100

j=i+1;

cn1=cn1+1;

Train(cn1,1:s2)=AA(i,1:s2);

Test(cn1,1:s2)=AA(j,1:s2);

end

TrainLong = Train;

%%%%%%%%%%%%%%%%%% PCA part (principal component analysis to reduce

%%%%%%%%%%%%%%%%%% number of features)

if(doPCA==1)

[coeffs, scores, latents] = pca(Train);

num_pca=NC; %This is optional but usually up to the number of images

evectors = coeffs(:, 1:num_pca);

Train2 = evectors' * Train';

Test2 = evectors' * Test';

Train = Train2';

Test = Test2';

newSize = size(Train);

s2 = newSize(2);

s1 = newSize(1);

end

newSizeI = size(Train);

s2 = newSizeI(2);

s1 = newSizeI(1);

XX=[Train; Test];

if(ZNormalize==1)

[XXX]=normalizemean(XX);

else

XXX = XX;

end

if(UseBias==1)

XXX(1:2*s1, s2) = 0.5; %bias term

end

TrainN=XXX(1:s1,1:s2);

TestN=XXX((s1+1):(2*s1), 1:s2);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Now all features are normalized to zero

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% mean and unity variance

%------------------------------------------------------

ssz=size(Train);

s1=ssz(1);

N = s1; %the number of training example

TrainD=Train; %to initialize, then TrainD will dynamically change

%Then we do the filtering process based on the RELIEF algorithm

%here we look at the training data and we look the whole binary code as

%a feature vector over the whole training data

%%%********************************************

%SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%% The Evaluation part %%%%%%%%%%%%%%%%%%%

%this is for the raw data recognition rate

ccntt=0;

for i=1:50

aAtom=Test(i,:);

aclass1=floor((i-1)/25)+1;

aDDD=dist(aAtom,Train'); %the Euclidean distance calculation

[ddd iii]=sort(aDDD); %this is to find the minimum distance

aclass2= floor((iii(1)-1)/25)+1;

if(aclass1==aclass2)

ccntt=ccntt+1;

end

end

disp('recog rate for test data using the features')

100*ccntt/50

%%%Now we see the category of the top HII using the original data

xCcntt(1:HII)=0;

for i=1:50 %over all the test data

Atom=Test(i,:);

class1=floor((i-1)/25)+1; %the class of the test pattern

DDD=dist(Atom,Train');

[ddd iii]=sort(DDD);

%iii(1:HII) = Xranking(i,1:HII);

for hii=1:HII

xClass2(hii) = floor((iii(hii)-1)/25)+1;

if(class1==xClass2(hii)) xCcntt(hii)=xCcntt(hii)+1;

xCCC(i,hii)=1;

else

xCCC(i,hii)=0;

end

end

end

100*xCcntt/50;

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

for j=1:50

xcorr=0;

for i=1:50

xcorr = xcorr + sum(xCCC(i,1:j));

end

xprec(j)=xcorr/(5*j);

xrecal(j)=xcorr/(5*25);

end

plot(xrecal,xprec);