Research On The Technology Of Automatic Marking System Based On Image Processing

Because of the efficient marking process,more objective and fair scoring mechanism,convenient management and many other advantages,the automatic marking system is gradually replacing the traditional manual marking method.Now the popular way used in automatic marking system to indentify the answer of objective questions automaticlly is Optical Character Recognition(OCR).This method has a higher requirements on paper's quality,and need to purchase expensive special identification equipment,so it is more suitable for large-scale tests.For the automatic marking system based on image processing appearing in recent years,it is also an identification for the objective questions based on the Rectangular box-filled mode.For this mode,the recognition rate relys too much on the quality of the the filling,which is easy to cause misjudgment,and it is not in line with the candidate's answer habits,and will take more filling time.For the problems existing in the filling mode,this paper presents an automatic marking system based on the handwritten letter recognition mode.It deals with the answer sheets separated from the questions to reduce the scanning workload and improve the speed of the image processing,save the running time and storage space overhead.The main contents of this paper are as follows:(1)According to the characteristics of the answer sheets image,simplified the tilt correction process.In order to solve the problem of large amount of calculation in the process of straight line detection with Hough transform,firstly,the edge of the target image area is detected,then the horizontal line points were screened.Finally,the Hough transform is used to obtain the tilt angle.(2)A method of bar code recognition based on vertical projection is proposed.The barcode image recognition technology is introduced into the information recognition process of the candidates,which simplifies the process of system identification and improves the recognition accuracy.The bar code recognition method based on vertical projection can realize the fast and accurate identification of the bar code images which are seriously polluted and incomplete.(3)A new method of handwritten-letter-feature extracting is proposed.Feature points extracted by traditional method of handwritten-letter-feature extracting are too many,resulting in a more complex structure of recognition system.An eight-point-feature extracting method combined statistical features with structural features is put forward according to the characteristics of handwritten letters.It is proved that the feature points extracted based on the eight-point-feature extracting method can accurately identify the characters on behalf of,which has a higher recognition accuracy.(4)Based on the eight-point-feature extracting method,the automatic recognition of handwritten letters is realized by the LVQ neural network optimized by genetic algorithm.For the reasong that the initial weights of the neural network are unreasonable,it may appear "dead" neurons in network.In order to avoid this case,the genetic algorithm is used to optimize the initial weights of the network,and the genetic algorithm is improved to avoid falling into the local optimal solution,also accelerate the speed of convergence process.The experimental results show that there are obvious improvement in the convergence and classification performance of the LVQ neural network optimized by the improved genetic algorithm.At the same time,the structure of the LVQ neural network based on the eight-point-feature extracting method is more simple,and the recognition accuracy is higher,which meets the performance requirements of automatic marking system.The methods and techniques studied in this paper have great significance for solving the key problems in the automatic marking system based on image processing,and are suitable for the marking work of small and medium-sized examinations.