Thread: image comparsion

hello
i am student and working on my final project which is related to image comparison and semantic image searching.i started image comparison on the bases of RGB but it is not a good technique when image is in different size or in different format.
i want to know which techniques can i use to get more efficiency in it and also for semantic image searching .

1 have to complete 2 task.1st to compare images and 2nd to show user other images which are related to user given image.
I m not finding any specific feature from image, just trying to show how similar two or more images on the base of color and feature.I m beginner in image processing so I started on the base of color.My algo is simple it gets the RGB value of every pixel from the images and then store separately in 2 arrays then compare them.if 2 images pixels values are totally similar then it show 100% otherwise show the percentage according to similar pixels.
My algo is going well for images of same size,, greysacle and also it giving efficient result if one image position is different(rotated,flip).

"Problem" is when both images are in different sizes or different format, it gives wrong result due to different pixels value.
Plz Tell me a better technique which can resolve my problem and also it compare on the base of features.

values are actually the combination of RGB code of some pixel like -1186128 (B=176 R= 237 G= 230)
I use 2 array and initialized size of array according to the size(height*width) of images then store RGB values according to pixels position.like pixel value on position (0,0) store in array index zero and so on.
After storing values of both images in different array,i simply used for loop to compare both arrays values.
totally similar mean both arrays are equal.

yes content of the arrays is an exact duplication of the RGB values of the image.
I don't understand what you want to say. I know my technique is not good but it is giving right result to some extent.

You are asking too much complicated question ,May be i am not giving you clear concept of my algo.
if all the pixel are not same then it just tell the number of similar pixels in both images.

Just a wild guess, but I suppose you could resize both images to a significantly smaller equal size using an averaging algorithm, and then go pixel by pixel getting the Manhattan distance (look it up) between each pixel, and then get the sum of the squares of these distances and add them up.

okk sir i will try to apply this..
Can you tell me which technique is good for semantic image searching using java??