Segmentation and Recognition of Chinese Bank Cheque Amounts

This dissertation, "Segmentation and Recognition of Chinese Bank Cheque Amounts" by Kwok-wai, Lau, 劉國偉, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: The Abstract Abstract of Thesis Entitled "Segmentation and Recognition of Chinese Bank Cheque Amounts" submitted by LAU Kwok Wai for the degree of Master of Philosophy at The University of Hong Kong in December 2003. Abstract A huge number of bank cheques drawn in Chinese are processed every day in Hong Kong, Mainland China, and elsewhere in Asia. This is an enormous and time-consuming task, which at present is still performed manually. Although much research has been done into automating the recognition of English and numeric amounts on bank cheques, it has so far proved to be very difficult to develop specific methods for recognizing Chinese amounts. The main problem is to correctly separate each character from the image of a character string in which the characters overlap with or touch each other. This study proposes three approaches for the recognition of handwritten Chinese amounts. They are the Extended Tseng's method, the Modified Yu's method and the Segmentation-free method. Each method has its own merits. The Extended Tseng's method merges character components by using knowledge-based merging method which significantly reduces The Abstract recognition errors because most of the merged components are complete characters rather than unknown fragments. The Modified Yu's method improves segmentation performance by accurately decomposing the connected segments whose width exceed pre-defined thresholds. The Segmentation-free method uses a sliding window to capture a sequence of frames from the string image, so that those character segments which have widths less than the preset threshold can also be successfully segmented. Of the three proposed methods, the Extended Tseng's method gives the best performance. With a test set of 192 bank cheques, it scored a recognition rate of 79.8%, a rejection rate of 16.6%, and a substitution error rate of 3.6%. The error rate of 3.6% is lower than that of other existing models, and is considered satisfactory. An abstract of 269 words DOI: 10.5353/th_b2776924 Subjects: Chinese character sets (Data processing) Pattern recognition systems Checks - Data processing