Best Adaptive Threshold Method for video image to text extraction

Question

I'm in search for best Adaptive Threshold method for image binarization. I have found "Adaptive Thresholding Using the Integral Image" by Derek Bradley and Gerhard Roth . Would you please help me to find out which Adaptive Threshold method is best suitable this time for binarizing documents having both text and other objects? Thanks in advance.

Answer 1

The common one used is Sauvola binarization (as cited in the paper above). However there has been a number of systems proposed that are meant to improve upon it, many specializing in specific areas, like the paper you mentioned above which seems to specialize in gradient lighting. Here is a paper that I came across a while back that may be what you are looking for. It discusses the binarization of text in video. The author has also provided code for both his approach and for the more standard Sauvola and Niblack methods. This could allow you to compare some results and see what suits you best.