i used the below code for image retrival through corel database but my results are not upto the mark,please verify my node and let me get your help to make necessary corrections,I use local color feature

tic
disp('query')
[filename, pathname]=uigetfile({'*.jpg'},'queryimage');
img=strcat(pathname,filename);
a=imread(img);
subplot(4,5,3)
imshow(img)
set(gcf, 'Units', 'Normalized', 'OuterPosition', [0 0.05 1 0.95]);
[r, c, numOfBands] = size(a);
%dividing image into 8x8 subblocks
bs=64;%block size
nob=(r/bs)*(c/bs);%Total no of blocks
k=0;
for i=1:(r/bs)
    for j=1:(c/bs)
        col1=bs*(j-1)+1;
        row1=bs*(i-1)+1;
        col2=bs*(j-1)+bs;
        row2=bs*(i-1)+bs;
       C{k+j}=imcrop(a,[col1 row1 col2 row2]);   
    end
    k=k+(r/bs);
end
for i=1:nob
D = rgb2hsv(C{i});
% split image into h, s & v planes
h = D(:, :, 1);
s = D(:, :, 2);
v = D(:, :, 3);
% Create the histogram quantized into 8×3×3 bins
X= zeros(8, 3, 3);
for k = 1 : numel(h)
  indexH = floor(8 * h(k)) + 1;
  indexS = floor(3 * s(k)) + 1;
  indexV = floor(3 * v(k)) + 1;
  if indexH > 8
    indexH = 8;
  end
  if indexS > 3
    indexS = 3;
    end
  if indexV > 3
    indexV = 3;
    end
    X(indexH, indexS, indexV) = X(indexH, indexS, indexV) + 1;
end
    x{i}=X;
end
disp('database')
CP=zeros(1,999);
str='.jpg';
for z=1:999 
b=num2str(z);
c=strcat(b,str);
a=imread(c);
[r, c, numOfBands] = size(a);
%dividing image into 8x8 subblocks
bs=64;%block size
nob=(r/bs)*(c/bs);%Total no of blocks
k=0;
for i=1:(r/bs)
    for j=1:(c/bs)
        col1=bs*(j-1)+1;
        row1=bs*(i-1)+1;
        col2=bs*(j-1)+bs;
        row2=bs*(i-1)+bs;
       C{k+j}=imcrop(a,[col1 row1 col2 row2]);   
    end
    k=k+(r/bs);
end
for i=1:nob
D = rgb2hsv(C{i});
% split image into h, s & v planes
h = D(:, :, 1);
s = D(:, :, 2);
v = D(:, :, 3);
% Create the histogram quantized into 8×3×3 bins
Y= zeros(8, 3, 3);
for k = 1 : numel(h)
  indexH = floor(8 * h(k)) + 1;
  indexS = floor(3 * s(k)) + 1;
  indexV = floor(3 * v(k)) + 1;
  if indexH > 8
    indexH = 8;
  end
  if indexS > 3
    indexS = 3;
    end
  if indexV > 3
    indexV = 3;
    end
    Y(indexH, indexS, indexV) = Y(indexH, indexS, indexV) + 1;
end
    y{i}=Y;
end
for i=1:nob
CP(z)=euclideanDistance(x{i}, y{i});
CP(z)=CP(z)+1;
end
end
B=sort(CP);
for z=1:999
     for M=1:12
if(CP(z)==B(M))
    subplot(5,3,M+3)
    z=num2str(z);
   imshow(strcat(z,str))
end
     end
 end
 disp('retrived')
  toc

[Benjamin Kraus]

You will need to format your code so that people can read it.

TUTORIAL: how to format your question with markup

[teja jayavarapu]

I think the format is clear in that file, if it is not up to the mark please notify me

[Benjamin Kraus]

Yes, you can read the code in the PDF, but you are more likely to get answers to your question if you update your question and indent your code properly. You can edit your question, and all you need to do is add two spaces at the beginning of each code line.

[teja jayavarapu]

I have written the below code for local colour histogram(where each image is subdivided into blocks and compare with the images in the database and find the most similar images.

[PROBLEM WITH MY CODE]

I am using corel database,i am thinking my logic with the below code was right but iam encountering some issues during execution,can you find out where i have done wrong with code.

screenshot(2)[for 523 image];screenshot(3)[for 9 image]

For 9th image it is showing some results.

But for 523(or)400 or some other images it is not showing correct results.

I THINK I HAVE MENTIONED DETAIL DESCRIPTION OF MY PROBLEM *

IF ANY INFORMATION NEEDED ABOVE THIS PLEASE LET ME KNOW*

[teja jayavarapu]

Please let me know if you could help me with this or not

[Walter Roberson]

It would be easier if you could attach the image files so that we could run the code to test with.

[teja jayavarapu]

Please let me know if you could help me with this or not

[Walter Roberson]

"Please let me know if you could help me with this or not"

No, you appear to need the solution faster than I can provide it, because I need to sleep at night. You had better hire a team of consultants.

[teja jayavarapu]

Sorry I didn't mean to bother you,I can wait for your solution, Hiring A team of consultants can give me a solution but completing the task under your guidance will give me complete idea.

Please count me in.I require your help.

Please reply with yes if you want to help me or else reply with no.

What ever May be your decision Thank You for all the you have made.

Answer 1

[teja jayavarapu]

http://wang.ist.psu.edu/docs/related/

This is the related link where the test images are available, I think you can manage to download images,sorry i can't upload the image set as it is above 5MB.In that link there test file of 1000 immages(.zip file).

[teja jayavarapu]

Please let me know if you could help me with this or not

Answer 2

[Walter Roberson]

• Your logic for indexing when you store the cropped images was not correct.
• You were failing to add in the pathname when you imread() and imshow()
• You were not pre-allocating arrays. This is a weakness rather than an outright error
• You were asking to imcrop by start and ending indices. imcrop() expects width and height, not ending indices.
• imcrop has a known weakness that is often overlooked: if you are not at the right edge then imcrop extracts one column more than it should, and if you are not at the bottom edge then imcrop extracts one row more than it should.

I have attached cleaned up code.

I am not able to test this further as routine euclideanDistance is not defined.

[teja jayavarapu]

You have made the necessary corrections(above mentioned errors) in the attached code or not

And sorry,I didn't attach euclidean distance. I will attach the function

[Walter Roberson]

The code I attached fixes the problems I described.

[teja jayavarapu]

I think it is taking much time to retrive the images (can you please check it again with my attached euclidean distance file

[teja jayavarapu]

I think there is a time difference in retrieving images (diffrent images taking different time)why this happens.have you checked the code with my Euclidean distance function

[Walter Roberson]

Your code

for i=1:nob
CP(z)=euclideanDistance(x{i}, y{i});
CP(z)=CP(z)+1;
end

overwrites CP(z) with euclideanDistance(x{i}, y{i}) each iteration of the for i loop. The result is going to be the same as if you had only executed the last iteration when i=nob .

I suspect you want to total the distances for the blocks, but it is also plausible that you instead want to add 1 each time the euclidean distance is less than a threshold (that is, count the similar blocks.)

Your code

for z=1:999
     for M=1:12
if(CP(z)==B(M))
    subplot(5,3,M+3)
    z=num2str(z);
   imshow(strcat(z,str))
end
     end
 end

tries to do a bit-for-bit equality comparison between two floating point numbers. You should avoid using == between floating point numbers. http://matlab.wikia.com/wiki/FAQ#Why_is_0.3_-_0.2_-_0.1_.28or_similar.29_not_equal_to_zero.3F

That display code does not refresh the display until all of the for z is done, so although images might be shown in positions, only the final result would be visible to users, after the code finishes and the program returns to the command prompt. But if you happened to be debugging, then because the debugger returns to the command prompt regularly, you would be seeing the updates as you debugged, so you would get a different impression of the output. Perhaps you want to add in an drawnow()

That display code does not always update any given subplot, so what shows up in any given subplot depends upon the most recent z that happened to update that position, so the output would be a mix of z values but the user would not be given any feedback as to which z it was. This is confusing.

It is not clear why 15 subplot outputs are given ??

... and I just detected that your splitting of each candidate image into blocks has the same imcrop problem as I fixed for the original image. That kind of difficulty could have been avoided if you had put the block-splitting code into a function called from both places.

Your block-splitting code could have been made simpler if you had used mat2cell.

row_blocks = floor(r / bs);
col_blocks = floor(c / bs);
used_rows = row_blocks * bs;
used_cols = col_blocks * bs;
C = cell2mat(a(1:used_rows, 1:used_cols, :), bs * ones(1,row_blocks), bs * ones(1,col_blocks), size(a,3) );
C = C(:);

This also avoids the weakness with imcrop() that I mentioned before.

[teja jayavarapu]

Yes you are right I am adding up all the distances of the blocks and comparing the resultant with the database. Is there is any problem with that.

[Walter Roberson]

Version that adds the distances attached: