Research On Form Recognition

Form plays an important role in people's daily work and life. It has various forms and a wide range of uses. With the popularity of paperless office, an accurate, efficient and strong form recognition system can greatly improve work efficiency, resulting in enormous economic benefits. Therefore, form recognition became the research hotspot of image processing.On the basis of studying and summing up the technology about form recognition, a set of deal flow of the form recognition is proposed, which include image preprocessing, image edge detection, mathematical morphology, form line extraction and form line fitting. Image preprocessing is the first step of form recognition, whose effect is directly related to the recognition accuracy. Because of the complexity of background noise in form documents and in the latter part of treatment are carried out on the local area, the improved method for image thresholding based on fuzzy index is to be used and get good results. Next step, according to least square method, skew correction algorithm based on least square method is proposed.After comparing several common used operators' performance on image edge detection, canny operator is selected to detect and extract image edge. Furthermore mathematical morphology is used to do some operation on image, such as corrosion, expansion, opening operation and closing operation, in order to enhance the accuracy of image edge extraction. According to the characteristics that most form lines are horizontal and vertical lines, mathematical morphology is used to extract the form lines, selecting appropriate structural elements to filtrate pseudo-lines generated by characters and noises. Therefore, it is no need to do preprocess including noise elimination and removing characters before form recognition. Table checking method is used to refine the extracted lines. Because refinement causes disconnected lines, least square method is used to fit lines.A lot of experiments are done for some algorithms used in above steps. The algorithms are compared for the effects and performance. The development of form recognition system comes true and the performance is good.