Research And Design Of Video Magnify System Based On Image Mosaic Algorithm

Electronic video magnify system is an instrument,using real-time image processing technology to improve the ability of reading of low vision people.However,traditional electronic video magnify system has a single function and usually uses a single camera to capture document image.Therefore,the acquisition angle is small and the image resolution is very low.In order to read the whole article,users have to move the device or target text while reading,so the user may have bad reading experiences.In view of this problem,this thesis first uses dual cameras to capture images and perform splicing to obtain wide-angle and high-resolution images.Then the layout analysis is used on the spliced images to separate the information of different paragraphs in the document image.Finally,the segmented image is OCR-recognized,and the function of distortion-free enlargement and discoloration of characters is realized.The main work is as follows:(1)Overall design of electronic video magnify system based on image stitching.This thesis compares different image stitching algorithms,layout analysis algorithms,and OCR algorithms to determine the design requirements of the algorithm.At the same time,the hardware and software environment of the system can be determined.(2)A document image stitching algorithm suitable for electronic video magnify system was proposed.In the image registration step,a region feature description method for document images is proposed,a single text region is extracted as a feature region,and its centroid is used as a feature point.The description of feature points is performed by describing the texture information of the feature region,and the feature matching efficiency is improved.In the image fusion step,a search algorithm of the optimal splicing seam of document image is proposed to eliminate the splicing ghost.At the same time,the multi-resolution fusion algorithm is applied for optimizing the image exposure differences.(3)The principle of layout analysis algorithm and OCR algorithm is studied.According to the problems of the algorithm applied to the electronic video magnify system,a complex text segmentation and optimization strategy is presented.The confidence level was used to analyze the layout analysis and OCR recognition results,eliminate error recognition,and improve the recognition accuracy.In addition,two methods of training data production and three training modes are introduced,which can be adapted to different situations of model training.(4)Based on Qt designed electronic video magnify GUI software.The software implements the video stitching function.This thesis tests the stitching performance of different resolutions.The test results show that the software can meet the real-time requirements.At the same time,the software implements the functions of single character segmentation,line layout segmentation,and paragraph layout segmentation.The OCR recognition function for different division modes is implemented.The distortion-free amplification and discoloration of the characters of the recognition result are realized.