A new visual signature for content-based indexing of low resolution documents

• Main Authors: Md Nor, Danial, Abd. Wahab, M. Helmy, M. Jenu, M. Zarar, Ogier, Jean-Marc
• Format: Article
• Language: English
• Published: 2012
• Subjects: T Technology (General)
• Online Access: http://eprints.uthm.edu.my/7097/1/J14168_5130d0b6fdee9bb0e61a4edec1d3837d.pdf; http://eprints.uthm.edu.my/7097/

This paper proposes a new visual signature for content –based indexing of low resolution documents. Camera Based Document Analysis and Recognition (CBDAR) has been established which deals with the textual information in scene images taken by low cost hand held devices like digital camera, cell phones, etc. A lot of applications like text translation, reading text for visually impaired and blind person, information retrieval from media document, e-learning, etc., can be built using the techniques developed in CBDAR domain. The proposed approach of extraction of textual information is composed of three steps: image segmentation, text localization and extraction, and Optical Character Recognition. First of all, for pre-processing the resolution of each image is checked for re-sampling to a common resolution format (720 X 540). Then, the final image is converted to grayscale and binarized using Otsu segmentation method for further processing. In addition, looking at the mean horizontal run length of both black and white pixels, the proper segmentation of foreground objects is checked. In the post-processing step, the text localizer validates the candidate text regions proposed by text detector. We have employed a connected component approach for text localization. The extracted text is then has been successfully recognized using ABBYY FineReader for OCR. Apart from OCR, we had created a novel feature vectors from textual information for Content-Based Image Retrieval (CBIR).