java - Feature/blob correlation and histogram analysis -

I am working on a sketch search engine that associates any sketching with a database of the database (DB Just about 40 photos are now) I am mostly doing this for fun, so I am not well aware of computer imaging techniques.

First, the rules of thumb are how to make histogram (bin size, ranges, etc.)? I'm using some histogram code (but ported to Javasvavi). Sometimes I get good results, sometimes I get bad results, most of the time I get "Meh" results. I am using a ton of bins with sizes and ranges and I am thinking that high-dimensional histograms can be compared to this answer.

Secondly, it seems that black makes me very strong presence in the current histogram setup (even a black dot change in whole result set). Should this be expected? Or did I screw up some? Example: and after dot: Note that I'm already getting pictures of" earth "in the form of" closed "matches.

I am also thinking about how I should use blob or feature analysis, I think things like SURF can be high because I only want to clearly compare blobs. I do not want to map templates properly. Any way I can compare the edge after passing through a Canny filter? (If possible, the following complexity):

For example Here, I want two smiley faces on top because the Sue Smiley "Blob" is more closely related to the size of the face while smiling than a bunch of obsession fruit or galaxy.

If you want to use your engine, go to (shameless plug, I know, I know - only works in FF / Chrome / Safari). Perhaps more experienced computer vision can give suggestions on the basis of people's results. Oh, I'm using the CV_COMP_BHATTACHARYYA distance while comparing the histogram (It appears that it gives the best results, though Che-Square is not bad).

Is there a background? this is important? Perhaps you need to see if a user has a background provided or not. So you should have 2 histograms per DB entry in "bus", with one BG, without one.

This will stop the Earth that appears with a dot like an apple.

Basic BG Separation, make a grand effort, then take "outside" and remove it from a copy of the original.