Efficient Segmentation Technique for Restoration of Degraded Images

Abstract

The process of extracting text from degraded document images is a very difficult task because of high inter/intra variations between document background and foreground. In this paper we propose a better document image binarization technique which can restore all kinds of degraded images and in this technique we use Adaptive image contrast. This Adaptive image contrast is a combination of local image contrast and local image gradient. To estimate colour in foreground and background we prefer local image contrast and to estimate objects in the picture, its magnitude, its direction we use local image gradient. In our project we first construct an Adaptive contrast map from local image contrast and local image gradient which is then binarized and combined with canny edge map by which we can identify text stroke edge pixels next its output is segmented by using local threshold segmentation based on intensities of detected text stroke edge pixels within a local window.Our project is very simple, efficient and involves minimum parameters using. This proposed method is applied onDIBCO 2009, 2011 and handwritten DIBCO 2010 and achieved accuracy of about 93.5%,87.8%and 92.03%.