Automatic Segmentation And Recognition Of Non - Fixed Format Printing Bills

With the rapid development of economy, Bank Notes application spread widely. Notes of automated processing is an important measure to improve efficiency,rigor,and security in the financial sector. However, the recognition of unfixed format negotiable instrument is still a technology bottleneck in Bank Notes recognition.This paper takes the unfixed format Bank Notes as the research object, using the theory of image processing and pattern recognition, etc, to construct an automatic identification system of the unfixed format negotiable instrument table, to extract and identify the table items information, such as ’trading institution’,’trading name’, etc.This system is mainly divided into four parts:The first part is the paper seal filter stage. We study a filtering method of red seal in this part, by detecting instrument seal area in the image, to enhance the characters covered by the seal and use the background gray to compensate seal area, to synthesize natural no seal image.The second part is the paper image segmentation stage. According to the different degree tilt in the actual photographs of negotiable instrument, we do the gray image tilt correction based on the different direction of projection. In This paper, we use the projection method to realize the ranks of segmentation of the paper, and then do the second line segmentation through a combination of direct projection and differential projection.The third part is the extraction of unfixed format paper list items. First of all, we divide the paper into half angle small pictures and identify them are whether character images or not. Then, by the colon number of whole paper, we determine the table and table value delimiter type. Finally, we get small pictures of the table and table value from different delimiters adopting different strategies.The fourth part is table item identification stage. We reposition the boundary of the small table image, shard angle and half-angle characters by the combining method of projection method and identify the feedback method, and then send the single word to corresponding classifier to be identified. The results of test show that the system has high performance, and it has reached the requirement of practical application.