Abstract

The digitization of historical documents in libraries is most important as they preserve the content and make the documents available through electronic media. The historical documents suffer from degradation due to damaged background, stained or damaged paper and similar other factors, so the results are not as desired. Binarization techniques can improve these recognition results. Here, the problem arises for selecting the correct threshold. The main goal or aim is to effectively binarize the document images that suffer from strain and smear, uneven background, holes and spots and various illumination effects by applying Adaptive Binarization Techniques. In this proposed work standard estimation Emean divided by 4 and Estd multiplying by three for evaluating the results. In post processing, area value less than 15 for connected components and threshold value 0.4 is used. These parameter values are estimated after analysing the results. It has been found that each technique has its own benefits and limitations. None of the technique is best for each and every case. Accuracy value is 96.8% for printed images and 98.4% for handwritten images of DIBCO 2009 dataset. The new proposed technique gives better results as compared with previous techniques in term of the performance evaluated by using Parameters such as PSNR, Precision, Recall, F-Measure, Accuracy, Sensitivity and Specificity.