N IMPROVED HYBRID BINARIZATION TECHNIQUE FOR DEGRADED DOCUMENT DIGITIZATION

In document analysis, binary image representation is essential format. In previous studies, different techniques are implemented and tried and true for different types of degradation in document image. But none of these methods are competent to properly ex tract the text from degraded background. However, few thresholding techniques are available but due to complex nature it cannot s olve all binarization problems. This paper evaluates a hybrid document image binarization technique which uses both local and global thresholding that divides the text from badly degraded document images accurately . The proposed method involves specialized evaluation technique for binarization to test samples from DIBCO 2009 - 13 dataset which have contras t parameter selection . The method uses both local image contrast and gradient to enhance the poor quality old script images and define edge pixels correctly. To evaluate the proposed work performance, we compare it with Otsu, Sauvola, Niblack, Bernsen and L MM method. The proposed method adapts all the binarization problems and brings highest 94.97% accuracy and gives superior performance compare to all other state - of - art techniques.