I would suggest histogram intersection or chi-square, as mentioned in https://github.com/liboyin/horc/issues/6. Not necessarily with SVM, though. An annoying bit would be what to do with the large, clean, white background. Maybe we should first filter out pixels with very high brightness?
I would suggest histogram intersection or chi-square, as mentioned in https://github.com/liboyin/horc/issues/6. Not necessarily with SVM, though. An annoying bit would be what to do with the large, clean, white background. Maybe we should first filter out pixels with very high brightness?