Design On Check Identification System Based On OCR

In recent years, check is used more and more widely, but identification of check is still mainly carried out manually, which is less efficient. Aiming at this problem, this paper mainly study the research of the check image recognition algorithms in the Check Automatic Identification System, thus realize the automatic identification of the check. Meanwhile, this paper makes an important application attempt of Optical Character Recognition Theory in the field of financial.First, this paper set up a hardware platform which is based on ARM and FPGA to scan and recognize image, and then preprocesses the image. The image acquisition sensor is calibrated in terms of exposure time and temperature compensation, this calibration can improve the uniformity of the image. The positioning and binaryzation processing of the image are conducted by using the least squares method and Otsu method respectively. As proved by experiments, the least squares method is an effective image location algorithm. And the Otsu method can retain the key information of image to the greatest extent.Based on the image preprocessing results, this paper proposes an iterative recognition policy for printed characters. The printed characters on the check are optimized based on the variable of binary threshold value, and automatic identified by combining with such character recognition algorithms as location, segmentation, feature extraction, etc. Then based on the neural network and classification theory, a two-stage classifier integration strategy for the recognition algorithm of handwritten financial Chinese characters is proposed. To realize the automatic recognition, the algorithm is based on the LDA feature extraction algorithm and was combined with the recognition algorithm of neural network classifier and KNN classifier. In order to enhance the reliability of the identification of check, this paper presents a two-stage classifier integration recognition algorithm for the magnetic character.Finally, all the recognition algorithms of Check Automatic Identification System are programmed and tested on the hardware platform which has set up in this paper. The results show that all of the identification strategies in this paper can effectively carry out the automatic identification of the check. The key technical indicators in the paper all meet the design requirements. And also the correctness and validity of the optimized theories proposed in this paper are also verified.