The Layout Analysis Of Postal Envelope And Character Segmentation

In this paper, we propose a postal envelope recognition system, and analyze the structure of it.Document analysis and Chinese character segmentation are two important parts of OCR system. The former segments the document image into several parts and distinguishes the parts text, image, drawing or table etc. We call them the parts blocks. The blocks got by document analysis will be different treated. The latter segments the text-line several parts including many indepent characters. Correct segmentation of text image directly affects recognition-rate of Chinese character.In our system, firstly, we pre-process envelope image, introduce basic method of two-value and noise's filtering accordingly, and present a way of deskew based on HHT algorithm. Secondly, on the basis of traditional document analysis technique, document segmentation and document understanding are introduced in detail. Then the thesis presents a probabilistic Viterbi algorithm to segment address block and gets the character. Finally, after independent Chinese characters are recognized and fitted with address database, we get the address.In this way, automatic recognition of envelope's address is realized. The presented algorithms have been integrated into Hanwang postal envelope recognition system.