Restoration of Noisy Document Images with an Efficient Bi-Level Adaptive Thresholding

An effective approach for extracting document images from a As an example, Otsu’s celebrated paper [8] chooses the optinoisy background is introduced. The entire scheme is divided into three submum threshold by maximizing the between-class variance with stechniques – the initial preprocessing operations for noise cluster tightening, an exhaustive search, while, Kittler and Illingworth’s work [5] introduction of a new thresholding method by maximizing the ratio of stanassumes two separate normal distributions and minimizes the dard deviations of the combined effect on the image to the sum of weighted error. In another study, Kapur et al. [6] have found the threshclasses and finally the image restoration phase by image binarization utilizold by maximizing the entropy of the gray-level histograms ing the proposed optimum threshold level. The proposed method is found of resulting classes. However, our proposed non-parametric to be efficient compared to the existing schemes in terms of computational scheme has assumed two different overlapping pixel clusters, complexity as well as speed with better noise rejection. one for background noise and the other for the image. Our Keywords— Document image extraction, Preprocessing, Ratio of stangoal is to restore the image cluster by, first, preprocessing the dard deviations, Bi-level adaptive thresholding. image to tighten the cluster distributions and then, searching further for the optimum bi-level threshold with a fast algorithm which basically maximizes the ratio of standard deviations of