Comparing Histograms In Search For Similar Images (java)

Asked Feb 21 '15 at 13:52

Active Feb 21 '15 at 13:52

Viewed 509 times

Okay so I have an original image and I generate a histogram with bins for it. When I hit search it iterates through a image database and generates their histograms and uses some distance measures to display in order the most similar images to the original image. This all works really well when the images have the same dimensions but not so well when the images have different dimensions. This doesn't work with images of different dimensions as the histogram will contain more or less values depending on if the image is larger or smaller throwing off the distance calculation.

So my question is is there any way to address this problem? can you normalise the histogram bins some way?

Thanks for reading through all that if you made it this far.

asked Feb 21 '15 at 13:52

Bad Dub

1,503
2
22
52

1

How about dividing each frequency of the historgram by the number of pixels in the image? – RealSkeptic Feb 21 '15 at 13:58
Why not simply rescale the images to the same dimensions? – runDOSrun Feb 21 '15 at 14:01
@RealSkeptic I was thinking that could work. Might give it a go now! – Bad Dub Feb 21 '15 at 14:26
@runDOSrun Would rescaling not lose some vital information though? – Bad Dub Feb 21 '15 at 14:26
@BadDub I think it depends on your images. In theory the quantities should change within each bin but the overall tonal distribution should remain when rescaled (you're basically just stretching the histogram on the y-axis). – runDOSrun Feb 21 '15 at 14:38

Comparing Histograms In Search For Similar Images (java)

0 Answers0