A New Contrast Based Degraded Document Image Binarization

Abstract

The binarization of badly degraded document images is very challenging job due to the presence of various degradations such as ink bleed through, stains, smear, blur, low contrast and nonuniform illumination. Many binarization techniques are proposed in the literature, most of these techniques are threshold based. This chapter proposes the binarization technique which uses the contrast feature to compute the threshold value with minimum parameter tuning. It computes the local contrast image using maximum and minimum pixel values in the neighbourhood. The high contrast text pixels in the image are detected using global binarization. Finally, the local thresholds are computed using high contrast image pixels within the local window to binarize the document image. It has been tested on benchmark datasets H-DIBCO-2010 and H-DIBCO-2016 in terms of F-measure, PSNR and NRM. The results are compared with the Bernsen’s and LMM contrast based binarization techniques and found to be outperforming these methods.