Handwritten Document Image Binarization: An Adaptive K-Means Based Approach

Degraded historical document images face many challenges in the process of optical character recognizing or word spotting, even after applying the traditional binarization techniques. In this paper, we propose a K-means based clustering technique for adaptive binarization of degraded document images. For validation of test results, we have used the recent dataset of Handwritten counterpart of Document Image Binarization Contest (H-DIBCO'16) comprising of highly degraded handwritten document images and computed detailed results of each image. In order to corroborate verification and validation, the experimental results are compared with three top winning ones in the contest and other prominent techniques in the literature. Experimental results reveal outstanding performance in the four evaluation measures compared with the top winners of the competition, claiming its effectiveness and validity conformance.