Research And Design Of Electronic Job Correction System Based On Machine Vision

With the Ministry of education to promote the curriculum reform, the OMR technology has been widely used in many large and middle school. But the OMR equipment can only read objective questions, it can not carry out the statistical analysis on all questions on the paper. This paper presented a scheme .which combined with the traditional paper form and electronic reading homework. In this scheme, the system could scan directly through the scanner paper gets the image file,then the image file could be processed. This information would be uploaded to the server, which could be checked and statistics by the teachers through the terminal.In order to complete the design of the image processing part and special paper as the goal, the image cutting system and the objective questions marking system were designed in this paper. This paper focused on the image preprocessing and image recognition technology used in image processing. This paper mainly finished the following contents:1.The design of special paper for the system: The working paper is rearranged,and the QR code was added to represent the student's personal information and other information. In addition, the positioning marks were set around for scanning and location.2.The image preprocessing: The noise, tilt and other negative factors in the scanned image should be corrected. Image feature extraction, segmentation and compression were also discussed. Based on the research of the related theories, library binarization, filtering and threshold segmentation were realized by using the OpenCV development library. Using the Hough transform and the corner detection of the positioning block, the tilt correction of the working paper is realized. Finally, the JPEG compression algorithm is used to compress the image file.3.The design of image recognition system: The number of digital identification ,the multiple-choice answers recognition and the QR code identification system were realized in this paper. By using the two methods of extracting the image pixel feature and Hu moment invariants, the function of the digital recognition was realized by constructing and training the BP neural network. A method of comprehensive detection and recognition based on gray value template matching and typesetting was presented ,which could identify the multiple-choice answers. The SDK of Zxing project of Google was used to realize the QR code recognition. The information of QR code in the paper provided the relevant information for the other two types of localization and recognition system, the three as an organic whole to realize information sharing, further improved the recognition accuracy.Through the test, the design of system conformed to the current domestic education situation. It was proved to be a practical and inexpensive solution for electronic homework correcting, which had a certain commercial value.