Research And Software Design Of Automatic Statistic Method Based On Image Processing

At present, a large number of survey questionnaires, statistical forms, goods declarations and other information are all input by hand, and it is an effective way to realize the job automation by means of computer systems to identify, gain and post process the information. Although there are some special software systems based on OCR technology in the fields of application, such as mail sorting system, bank bill analysis system, ballot automatic statistic system and so on, the studies on the automatic treatment system for general data forms based on the constraints without fixed forms or special limits for filling are still very rare, and some relevant technical problems also need to be addressed.This paper has its background of the project research and development of "Construction of Popular Science Education System and Statistics of Popular Science Resources Based on Electronic Map", mainly researches the main support technologies in automatic form recognition and statistics, including definition and description model of general forms, handwritten element check box, segmentation of digits and recognition methods, aims at building up a system to automatic identify, analyze and count according to user-defined forms. For the form description problems, a general priori knowledge description model and a solution to define layout by XML document are proposed. Through studying the linear grating conversion algorithm, a multi-point fast linear scanning conversion algorithm is provided, and on the basis of its recurrence relations, a method is established to fast test and correct skew images based on multi-point recurrence and using two shears instead of rotation. Only relying on integer operations, can this method correct the images inclined at a low angle, without calculating inclination angle as well as rotation transformation. Owing to the feature that there are some differences in areas between the check boxes and words in common printing system, a new check box fitting line-block segmentation and area features of check boxes and an algorithm to recognize and identify whether the check box is chosen are constructed, in order to solve the problems such as uncompleted check boxes, and word confusion. The written digit string is divided by upper and lower boundary projection and linear segmentation algorithm. Rough grid and intersection features are combined to identify the written digitals by BP neural network algorithm.The experiments and analysis based on these algorithms show that they are effective to resolve the problems which are in the main steps of construction system. Among them, fast skew correction algorithm has relatively high precision and efficiency when the angle is low, while the recognition and identification algorithm of check box has its advantages in dealing with broken and confusing words. The automatic analysis statistic system is initially established by using of above results to allow user-defined form structure, and to support written digits as well as check box contents.