Some experiments on variable thresholding

Abstract Chow and Kaneko proposed a method of variable thresholding in which an image is divided into windows; thresholds are selected for those windows that have bimodal histograms; and these thresholds are interpolated to define a variable threshold for the entire image. This method was applied to several TV images of machine parts; the results obtained appeared to be considerably better than the results of thresholding at a fixed level. An extension of the method was defined that allowed histograms to be either bimodal or trimodal; this yielded some further improvement in the results, but was also more sensitive to shadows. Finally, an adaptive quantization scheme, based on histogram peak sharpening, was applied to two of the images; the results do not seem to be as good as those obtained using variable thresholding.