Design And Development Of Intelligent Assistant Marking System Based On Computer Vision

With its rapid batch processing,objective and fair grading,and more intuitive and convenient management,intelligent examination papers have become the main development direction of examination papers today.At present,the widely used automatic examination method is mainly based on the cursor reader and the answer card.This method has high requirements on the quality of the answer card,requires expensive special equipment,and is difficult to maintain in the later stage.This method is difficult to be widely used in small and medium-sized examinations.In response to some problems in the existing automatic examination papers,and with the rapid development and a large number of applications of computer vision and artificial intelligence technology,this paper designs and develops an intelligent examination paper system based on computer vision.Recognizing the test papers that students have answered directly has high application value.The main research contents of this paper include image pre-processing of test paper,detection of relevant content of test paper layout,information recognition of test paper target area and design and implementation of system.First of all,it is necessary to solve the problem of noise and tilt caused by certain factors in the test paper image scanned by the scanner.By analyzing the characteristics of the test paper image,the image is grayed and binarized,and then the average filter is used to image After denoising,and then using Radon transform to perform tilt correction,a clearer image of untilted test paper is finally obtained,which lays a good foundation for subsequent test paper recognition.In the relevant content detection part of the test paper layout,the content that needs to be detected includes QR codes,questions and handwritten texts.Through research and related research,the following schemes are designed: First,collect and produce a test paper data set suitable for the needs of this article.QR code,using the optimized Faster R-CNN target detection algorithm for detection;for the problem detection on the test paper,because the problem type is more diverse and the content content contains complex forms,there is no more unified model.The detection algorithm CTPN(Connectionist Text Proposal Network)detects the title part of the "title number + text",and then designs the title segmentation algorithm according to the coordinates of the upper left corner of the detection frame,which better achieves the division of the title area;for handwritten text detection,in order to solve the handwriting The problem of confusion with printed objects is easy to detect.In this paper,inspired by the idea of fine-grained image classification,a bilinear pooling module is designed to improve the CTPN algorithm,which improves the accuracy of handwritten text detection.The information identification part of the target area of the test paper mainly includes QR code decoding and text recognition.The QR code decoding uses the more mature ZXing open source library to accurately identify the information in the QR code.The text recognition scheme is designed as follows: First,a convolutional neural network with three residual blocks is designed to extract the spatial features of the text image,and then the designed two-layer BLSTM(Bidirectional Long Short-Term Memory)is used to fuse the feature vectors and extract the text The sequence characteristics of the text are finally transcribed and translated through the CTC(Connectionist Temporal Classification)algorithm to obtain the text sequence.Based on the above research,use the Tensorflow deep learning framework to train and export related models,and use the C # programming language and winform form program for client interface development under the Windows system,and use the My SQL database tool to save user data information and student exam information.The final design integration results in a complete intelligent examination assistant system.