The Key Technology Research And Realization Of Examination Paper Recognition

The automatic identification of the test papers' content is the key technology for the informatization of the examination content,which is the important part of digitizing the print-out test papers.This paper takes the image of the test papers as the research object andfocus on the research of some key technologies,such as image denoising,image correction,layout segmentation,handwritten digit recognition and so on.(1)A denoising method based on morphology for the image of test papers is proposed and implemented.According to the large difference of the gray value between the foreground and the background in the image of test papers,the two parts of the texture and the noise are distinguished from the background by the image binaryzation.According to the regular gray structure of the test papers' image,a Morphological filters is designed to further distinguish between texture and noise,as well as preserve the texture details using the method of connected domain partitioning.The experimental results show that the denoising method proposed in this paper has better visual effects than the traditional method on the examination papers' image and is closest to the original image under the mean square error criterion.(2)A novel method of test papers' skew Correction based on Mark points is designed and implemented.A set of mark points for geometric correction is designed according to the layout characteristics of the test papers' image.In the mark points positioning,this paper first coarsely locating the area of the mark point because of the attributes of the test papers' image with obvious boundaries.In addition,contour tracking technology is used to get contour description information of the candidate mark point.Finally,the true mark points are screened through the features of shape,duty ratio and size.Using the geometric relation of mark points to judge the upside down of the test papers' image and calculate the slant angle.According to the slant angle,a fast rotation algorithm based on improved DDA is adopted to correct the image.Experimental results show that the proposed method can quickly and effectively achieve the slant correction of test papers' images.(3)A layout segmentation algorithm based on segmentation line composition is studied and implemented.Because the test papers' layout boundary is almost the regular rectangle,this article uses the run-length encoding to screen the candidate separation line.According to the characteristic that the line to be selected in the same direction has a clear line cluster,the hierarchical division algorithm is adopted for the selected dividing line to obtain the divided rectangular blocks in the horizontal and vertical directions,and the rectangular middle line is taken as the dividing line of the layout.The undirected graph is constructed on the basis of layout division line,and the layout area is searched for in the undirected graph by the minimum round algorithm.In this paper,a layout segmentation experiment is carried out for different layouts of the test papers.The results show that the algorithm can effectively separate layouts of the test papers.(4)A digital recognition framework of fuzzy handwriting based on convolutional neural network is proposed and implemented.Using Lenet-5 to solve the recognition problem of students ' handwritten Student No and teachers' handwritten scores in the test papers' images.If the handwritten numbers are too small in the test paper,the image will be blurred after being normalized.In order to reduce the character recognition rate caused by image ambiguity,this paper combines image enhancement network and handwritten digital recognition network to form a fuzzy image recognition framework and uses the model training strategy proposed in this paper to carry out parameter training.Experimental results show that the framework proposed in this paper can effectively improve the quality of the blurred image before handwritten numeral recognition and effectively increase the recognition rate of handwritten digits in the blurred image.(5)A test papers' recognition system based on the research of this paper is introduced.At last,analyzing the hardware requirements and design points of the test papers' recognition system and introducing the main functional modules.