According to the feature of numeral image of electronic balance photographed by CCD, MATLAB, especially its video image toolbox, was used to achieve the manual character region location, automatical character segmentation and automatical digit recognition of the digit image in video signals. The method is able to recognize the values in the video signals with more than 97% correctness in multiple sampling tests. However, there were also some abnormal data caused by LED light retention in the dynamic display process and recorded by high frame rate video. The problem was solved by replacing the points outside the range with neighborhood median. This method is simple, highly adaptable and reliable, thus facilitate the research on filter media surface wettability and provide a set of reference for solution this kind of research.