Text Reconstruction using Torn Document Mosaicing

Abstract

Text extraction from torn documents is a critical process in the field of document reconstruction. Estimation of fragment orientation based on the text orientation is proposed to get an exact rotational angle with x axis. Corners detection with extraction of the pick points is introduced as a basic feature for the boundary matching process. A novel approach of single matched corner identification for finding the corresponding points between two oriented fragments is implemented. Dilation process is used to extract the background of the text from the text image. Removal of background from text image is devised for the clear visualization of seamless text with black or white background which is responsible for making the merging process easy. Shift and Merge method is implemented for positioning of the matched fragments. A function called as a maskblending is utilized to merge the extracted texts from the shifted fragments along masked irregular shape. An experimental result clearly visualizes the different steps of document mosaicing.