Offline Handwritten Body Text Recognition Research

There are mainly two kinds of Chinese character recognition entitled on-line and off-line. The on-line Chinese character recognition system, which accepts on-line inputs, utilizes a great amount of information such as the stroke order; write pressure and other information, while the off-line Chinese character recognition does not have such information so it faces a little difficulty. And handwritten characters vary not only in writers but also in fort and size, so how to solve these issues becomes the focus of research.This paper adopts the method of two level classfications to carry out recognition, by which we put forward a kind of method based on unit circle projection to class approximately and use texture feature based on gray level run length matrix to class accurately and gives the last output. We take the formal character of Chinese numbers in bank bill as recognition sample for experiments, and the results are satisfying. In the aspect of the noise handling, we give an adaptive threshold segmentation algorithm based on histogram besides threshold segmentation method and filtering arithmetic. We use Mathematical morphology principle and one direction sharpening algorithm to solve the problems which occur in noise extraction.Character extraction and standardization are the important part of character recognition. In this thesis, we give a simple method for character extraction, and then standardize the character based on the center of gravity. At last, we use Mathematical morphology principle to thin the character.The primary work of this thesis includes giving an adaptive segmentation algorithm, and using morphology principle and sharpening algorithm to solve the problems occurring in noise removing; illustrating a simple method for character extraction; demonstrating the difference between serial thinning method and parallel thinning method; giving a algorithm based on unit circle projection and combining the texture feature for recognition. The character recognition software is developed in C++ Builder environment.