Automatic Scoring System Based On Image Processing

Although with the improvement of the social education system and the specification of examination system, the types and ways of examinations are also growing, the traditional paper-based test is still the mainstream of the current examination system. Optical mark reader and marking system based on image processing are the main current automatic marking systems. The two marking systems are expensive, have high demand for paper and printing, have more restrictions on the format of the answer cards, so they are difficult in application.. In this paper, in order to make up for the inadequacy of the current exam systems, we developed an automatic recognition software which can be used to support a variety of answer cards and be compatible with various paper and printing methods. This paper focuses on the design and implementation of automatic identification software, and the key technology and solution of image processing. Basically, the paper includes the following main aspects:1)Sheet classification and card format definition software: Through the investigation of a large number of answer card styles, we classified all the cards into four different types, and designed a card format definition software according to the characteristics of different types. Users can generate a particular XML file which can store card information with the card format definition software so as to recognize each kind of cards in a batch.2) Precise positioning of the mark points to be identified: Precise location of the mark points is the premise and foundation of the accurate identification of answers. We proposed different answer sheet types to locate different target points. For type 1 and type 2, we used sync-point to locate mark points precisely; for type 3 and 4, we proposed a plan which based on locating the relative position of benchmark point. A great deal of experimental results proved the positioning effect was very satisfactory.3) The recognition of the mark points: In the process of image acquisition, the range of gray-level and contrast may be different among different answer sheets because of the various qualities of pencils, the varying filling areas and pressure. Therefore, we propose a recognition algorithm combined with the mark points’ grayscale, areas and consistency of all sheets. The results showed that the algorithm has a good performance on recognizing the answer sheets with different shades and areas.The project is developed based on C++ and Open Cv. With fast identification speed and high accuracy, the software can be compatible with most of the answer sheets used in current tests,and has good recognition performance for the cards whose mark area is filled irregular or in different shades.For now, our software has been successfully applied to the college entrance examination, the postgraduate exam, College English Test Band 4 and Band 6, senior high school entrance examination, higher self-examination and other large tests. Meanwhile, it is used in some universities and schools for daily examination grading work, and has achieved good results.