Touching character segmentation for printed Odia document
Sanjibani Sudha Pattanayak, Ramesh Chandra Malik, Sateesh Kumar Pradhan, Aradhana Kar · 2021
Character segmentation is one of the most challenging tasks in Optical character recognition. The main aim of this paper is to develop an algorithm for touching character segmentation. Here an ensemble approach is introduced for segmenting the characters from the words of a printed Odia document. It integrates projection-profiles based method, connected component-based method, and a novel algorithm for touching characters. The projection-profile based method segments the characters which are well-separated by space. The segmented components that we get from the projection profile-based method are used to determine the approximate width of the character. The connected component-based method is being used for the segmentation of the overlapped characters. The remaining filtered word images contain touching characters. For the touching characters, a novel algorithm has been developed by using the average width of a character. The proposed ensemble strategy has been evaluated to identify the archival documents as well as new documents.