Image Reconstruction of Old Document Images Using Compound of Binarization and Recursive Connected Component Labeling

This paper proposes a recursive related component labelling method (PKTR) for reconstructing document images that connect two disconnected or unrelated characters in a binary document image. PKTR is based on the Pixel Window Tracking Widening (PPTP) subprocessor on binary images and the Neighborhood Distance Tracking (CBC) subprocessor. The PPTP sub-processing method describes pixel-window tracking in seed image frame pixel searches and the frame pixels of authentic differentiation images. The CBC subprocessing method fills the intervals of the selected characters in the binary images based on the numbers marked in the tracking process on the outline image. Two image-edgebased contrast assessment techniques, EBCM and EBCM1, were used to evaluate the completeness of the edges of the PPTP method images with the best results in the DIBCO Binary Image Document Competition (DIBCO) 2009, 2010, 2011, 2012, 2013, and DIBCO 2014 datasets compared to the minimax method (image edge floating sensitivity modification method), new methods of double threshold, and ant colony optimization methods. Performance of the PPTP method was achieved, namely EBCM: 375.65, 224.07, 185.16, 87.81, 99.45, 213.76; and EBCM1: 0.92, 0.55, 0.46, 0.22, 0.24, and 0.53. This shows that the PPTP method was better at reconstructing images of old documents than the other methods.