Digital Scoring Image Recognition Based On SVM Classification Design

With the development of educational technology, presently, traditional OMR scoringsystem is frequently applied in many large-scale examinations. But in such a scoring system,there exist the following obvious deficiencies: enormous investment, strict requirements inproducing answer sheets, complexity in operation, and inability to save the images. This studyaims to establish an image recognition technology of high precision and efficiency which issuitable for both scanner and camera marking so as to overcome the above-mentioneddeficiencies, in the hope of developing a certain type of digital image automatic scoringsystem of low cost, easy to operate and more suitable for individual use.This paper shows analysis and research of the processing of answer sheets, the featureextraction and optimization and the SVM sorting algorithm. On the basis of the aforesaidresearch and experiments, an intelligent digital scoring system applied in examination systemis designed and realized, using the programming language Visual C++. The test results provedto be of practical use.This research’s innovation is listed as follows:1、By using nonlinear geometric correction algorithm. The algorithm can be completed atthe same time process image tilt correction and normalized, effect of perspective distortionimage, the correction speed is fast. proposed a simple and practical answer sheet imagecorrection method, Marking system for image processing technology was brought to theimprovements.2、Based on the answer sheet image feature extraction and optimization algorithm design.We extract its geometric feature, histogram statistical feature and texture feature according tothe characteristics of the answer sheet image. Then genetic algorithm (GA) is adopted tooptimize the image features and reduce the feature vectors to8components. The fitnessfunction is a class of interval value divided by the type of pitch value.3、Based on the image recognition of SVM sorting design, using the image feature data,numerous simulation experiments are carried out under the condition of different SVM kernel function and parameters. According to the comparison of experimental results, theexperiments show a good sorting effect and meet the needs of actual practice if using RBFkernel function.The precision of system identification has reached97%, different types of questions andscores can be designed according to specific needs. In the mean time, answers recognition andstatistics analysis can be finished automatically. If applied to the use of questionnaire, onlysome small modifications of the system are required. The automatically formed statisticalinformation brings its users great convenience. Although the system recognition error rate isvery low, but if the error occurs in the Student ID area, will cause the number of Student IDmistake of marking, is very difficult to deal with the errors.