Decompose-threshold approach to handwriting extraction in degraded historical document images

Historical documents contain important and interesting information. A number of techniques have previously been proposed for thresholding document images. In this paper a new thresholding structure called the decompose-threshold approach is proposed and compared against some existing global and local algorithms. The proposed approach is a local adaptive analysis method, which uses local feature vectors to find the best approach for thresholding a local area. Appropriate algorithm(s) are selected or combined automatically for specific types of document image under investigation. The original image is recursively broken down into sub-regions using quad-trees until an appropriate thresholding method can be applied to each of the sub-region. The algorithm has been evaluated by testing on 10 historical images obtained from the Library of Congress. Evaluation of the performance using ´recall´ value demonstrates that the approach outperforms any existing single methods.