Normalized SAD Method for Chinese Document Image Registration

Abstract

In the area of digital content technologies, Chinese document image registration is an important task of document mosaic system, also an important research content of distorted document recognition system. The Sum of Absolute Difference, SAD, is a similarity criterion for image registration. To get better document image registration accuracy, a document image registration method based on normalized SAD is proposed. In this method, a pre-processing procedure is performed to identify the inflection point for the normalized SAD registration. Then gray histogram of the matched image is normalized firstly so that the different laminations of the two document image can be eliminated. Then the two document images are registered by the SAD method. At last, a post-processing of the normalized SAD registration method is performed to the two registered document images so that the two distorted document images can be mosaiced together and get a flatter document image. Experimental results show that the document image registration method based on normalized SAD is more robust than the classical SAD method and the Optical Character Recognition (OCR) rate of the new mosaiced image is markedly higher than the two original distorted document images.