Simple and Efficient Method of Low-Contrast Grayscale Image Binarization

This paper describes a simple analytic method for determining an optimal threshold position during low-contrast grayscale image binarization. The described method uses no tunable or user-set parameters and is invariant to both image size and pixel intensity value range. It uses both the pixel intensity histogram of the analysed image and its first-order derivative to estimate the threshold position. There is an assumption that the histogram is bimodal by its nature, however, the method can cope with the histograms where the peaks are very close or even overlapping due to low contrast. The method performance is comparable with the traditionally used Otsu's method [1] on unproblematic images, however, it significantly outperforms Otsu's on specific images, like e.g. electron microscopy of composite material fibres or x-ray mammography, where the bimodal histogram is almost (but not entirely) collapsed into unimodal.