Soft Thresholding for Visual Image Enhancement

AbstractThresholding converts a greyscale image into a binary image, and is thus often a necessary segmentationstep in image processing. For a human viewer however, thresholding usually has a negative impact on thelegibility of document images. This report describes a simple method for “smearing out” the threshold andtransforming the greyscale image into a different greyscale image. The method is similar to fuzzy thresholding,but is discussed here in the simpler context of greyscale transformations and, unlike fuzzy thresholding, itis independent from the method for finding the threshold. A simple formula is presented for automaticallydetermining the width of the threshold spread. The method can be used, e.g., for enhancing images for thepresentation in online facsimile repositories. 1 Introduction Thresholding can be considered as a special case ofimage segmentation: it partitions the image pixels ofa greyscale image into foreground (typically “black”)and background (“white”) pixels, thereby transform-ing the greyscale image into a binary image. As it isboth an essential and a possibly difficult preprocess-ing step in many image processing systems, in par-ticular for document image recognition, many differ-ent thresholding techniques have been proposed in theliterature [1] [2]. The thresholding algorithm itself isvery simple: let f(x;y) be the grey value of the imageat pixel position (x;y); then thresholding with thresh-old ttransforms this image into a binary image f~(x;y)as follows:f~(x;y) =