Code covered by the BSD License  

Highlights from
K means clustering for Image Compression

Be the first to rate this file! 35 Downloads (last 30 days) File Size: 4.99 KB File ID: #36376

K means clustering for Image Compression

by

 

25 Apr 2012 (Updated )

K-means clustering is a popular vector quantization method for data compression.

| Watch this File

File Information
Description

k-means clustering is used for Image Compression. In this matlab program, the feature vectors are simply the N X N non-overlapping blocks of pixels in the image. Like a scalar quantizer, a vector quantizer has a quantization levels called codevectors and the set of K such codevectors is called codebook of size K.
    
K-means clustering is an iterative process in which the codevectors are refined every stage by computing the centroid of the input vectors which belong to the respective cluster.

Required Products Image Processing Toolbox
MATLAB Compiler
MATLAB Coder
MATLAB
MATLAB release MATLAB 7.10 (R2010a)
Tags for This File   Please login to tag files.
Please login to add a comment or rating.
Comments and Ratings (8)
31 Mar 2014 Mayuri

I want to compress a gray-scale image using vector quantization. I have not understood how you have generated the codebook. Will you please explain me the steps for generation of the same in your code

14 Mar 2013 Edwin

Hello, thanks for your sharing.

However, when I tried to test this code, i got some errors pposted below...my input image is a 181*217 png image...would you please help me to solve this problem?

Thanks for your help in advanced!

Attempted to access I1(1,2.41421); index must be a positive integer or logical.

Error in ==> K_MEANS_PREPROCESSING at 18
s((L^0.5*size(I1,2))*(i-1)+L*(j-1)+(L^0.5)*(k-1)+l)=I1(sqrt(L)*(i-1)+k, sqrt(L)*(j-1)+l);

Error in ==> Image_kmeans_main at 72
I_re=K_MEANS_PREPROCESSING(Img1,L,K);

08 Jul 2012 Vinay Kumar Tadepalli

Hey Nidhi, Thanks for all the comments. I have updated the file again and this time i have tested for different images and it is working fine. You need to supply input parameters such as L & K for the program. First try with L=4 and K=16. This will execute in approx. 50 sec. Then you can try for other combinations.

07 Jul 2012 Nidhi

Thanks for your support. This new code doesnot give the error I got earlier, but the program is busy in its operation and finally it says "elapsed time". The program works for about 21 min approx.
Hope you will help me in resolving this issue.

Thanks,

Nidhi

07 Jul 2012 Neha

resolved with the above query.. m new to MATLAB.. i need k mean clustering technique to apply on my images for my project having deadline 15th july.. plzz help me for classifying my 255*255 grayscale satellite images into 3 clusters..

will be very grateful to u..
looking for possitive feedback in anticipation...

06 Jul 2012 Nidhi

In fact I tried using the original code with a 512*512 image but it too gets hanged or matlab shows status as busy and then nothing happens.

Thanks

Nidhi

02 Jul 2012 Vinay Kumar Tadepalli

Hey, Thanks for the feedback. Actually, when i wrote the code, i have used 512 X 512 image and didnt generalize it for all dimensions. I am pasting a new code here which should work for all dimensions of images.
---------------------------------------
clc;
clear all;
tic;
I1=imread('C:\Users\as\Documents\MATLAB\Test Images\1.gif');
%% Creating L-dimension vectors from the Image.
% For an Image, L is an integer power of 2.
s=zeros(1,size(I1,1)*size(I1,2));
L=4;
for i=1:size(I1,1)/sqrt(L)
for j=1:size(I1,2)/sqrt(L)
for k=1:sqrt(L)
for l=1:sqrt(L)
s((L^0.5*size(I1,2))*(i-1)+L*(j-1)+(L^0.5)*(k-1)+l)=I1(sqrt(L)*(i-1)+k, sqrt(L)*(j-1)+l);
end
end
end
end
imshow(I1);
K=256; % K is the Codebook Size.
[codebook,ClusterNum]=kmeans_clustering(s,L,K);
% Reconstructing the Image using the Codebook vectors (Cluster Centres).
s_re=zeros(1,length(ClusterNum)*L);
for i=1:length(ClusterNum)
s_re(L*(i-1)+1:L*i)=codebook(ClusterNum(i)).samples;
end
I_re=zeros(size(I1));
for i=1:size(I1,1)/sqrt(L)
for j=1:size(I1,2)/sqrt(L)
for k=1:sqrt(L)
for l=1:sqrt(L)
I_re(sqrt(L)*(i-1)+k, sqrt(L)*(j-1)+l)=s_re((L^0.5*size(I1,2))*(i-1)+L*(j-1)+(L^0.5)*(k-1)+l);
end
end
end
end
clear s s_re;
figure, imshow(uint8(I_re));
toc;

01 Jul 2012 Nidhi

When I use a .jpg file it says :
"??? Attempted to access I1(1,247); index out of bounds because size(I1)=[185,246].

Error in ==> image_kmeans at 17
s((L^0.5*512)*(i-1)+L*(j-1)+(L^0.5)*(k-1)+l)=I1(sqrt(L)*(i-1)+k,
sqrt(L)*(j-1)+l); "

I tried changing the size of the image but still it gives the same error. Only the coordinate or pixel value of out of bound changes.
Can you pl. tell me the size of image which I should use.
Thanks,
Nidhi

Updates
02 Jul 2012

The file I uploaded previously will fail to work for Images of different dimensions, since it was not generalized for all the dimensions. This updated file contains the instructions for working on different dimensions of Images.

09 Jul 2012

This file solves the issues with different dimensions of the images and also prints the output in a user-friendly manner.

15 Mar 2013

I have revamped the code completely to make it more efficient and hence run faster. The bugs in the previous version are fixed and hopefully its bug-free now. See ya!! :)

15 Mar 2013

I have revamped the code completely to make it more efficient(run Faster). I have rectified all the previous bugs. Feel free to contact me at my email address. See ya!!

22 Apr 2013

A small bug is fixed.

Contact us