Design And Implementation Of Image Matching System Based On OpenCv

With the development of China's society and the advancement of modernization,rural modernization has been implemented in every household,among which the reform of toilet facilities is an important part of the construction.Therefore,a Hebei science and technology company has developed a "Rural Toilet Revolution Management System" based on rural toilet reconstruction.Its main function is to assist the government to supervise toilet reconstruction.At present,the system has achieved complete functions in terms of farmers uploading images and government inspection images.However,some farmers falsely upload their own toilet images in order to obtain subsidies,resulting in duplicate and false toilet images in the system.The images newly uploaded by subsequent farmers can not be repeated detected with the stored image sets in the system,which leads to the system's inability to provide the government with real and effective images of toilet reconstruction,which brings difficulties to the smooth implementation of the toilet reconstruction plan.Therefore,the authenticity detection of the images in the system that have completed the toilet renovation and the assurance of the authenticity of the images newly uploaded by the subsequent farmers have become a problem that needs to be solved urgently.In this thesis,after studying a large number of image matching technology related data,on the basis of "rural toilet revolution management system" to preprocess the system image,and select one of the five thousand toilet images have been transformed as the data object to build image matching model.In the process of building the image matching model,the ORB algorithm's FAST corner extraction and BRIEF feature description were improved and optimized to improve the matching speed and accuracy of the system.After a violent match,the optimal similarity threshold is calculated through the Logistic regression model,which is used as the criterion of this system,and finally the image matching system is realized.The image matching system matches the stored images in the "Rural Toilet Revolution Management System" and the newly uploaded images for similarity,and uses a secondary manual check method to review,so that the system's duplicate check accuracy rate reaches99.6%.This thesis adds a duplicate check function to the "Rural Toilet Revolutionary Management System" to provide authenticity assurance.The main work and innovations are as follows:(1)Construct an image matching system.Based on programming languages such as Open CV and Python,the image matching system is realized through ORB algorithm.Solve the problem of some farmers uploading false and repeated images,realize the duplication check function,ensure the authenticity of the images uploaded by farmers,and provide guarantee for the follow-up department to supervise the toilet renovation.(2)Optimize the process of feature point extraction.On the basis of ORB algorithm,FAST corner extraction is carried out on the image.In order to improve the anti-jamming ability of the algorithm,the adaptive threshold of different images is first calculated,construct a quad-tree structure to divide the image,and select the appropriate corner point from each node as the feature point for subsequent feature matching.Successfully solved the problem of reduced matching accuracy due to too dense corners,and at the same time improved the algorithm's anti-interference ability.(3)Realize image rotation invariance and expansion invariance.Firstly,the image was converted to five different levels based on the image pyramid method.Then,the direction vectors of feature points and centroid were constructed using the gray centroid method.Then,the R-BRIEF description algorithm was proposed to describe the feature points,and the feature descriptors with rotation invariance and scale invariance were obtained.Successfully solved the problem that some farmers would falsely upload photos taken at the same place at different angles or at different distances from the same place.(4)The feature matching was completed and the optimal similarity threshold was selected.Firstly,the similarity ratio between feature descriptors was calculated by Brute Force,and then the similarity between different images was obtained based on the Hamming distance.The Logistic regression model is further used to calculate the optimal similarity threshold,and this threshold is used as the evaluation benchmark of the system to achieve repeatability detection between different images,which successfully improves the evaluation accuracy of the system.(5)Create image feature information library.The feature descriptor information of the image is stored in the image feature information database,and the subsequent image matching can be directly matched with the stored feature information in the image feature information database without any operation of feature extraction on the stored image in the system.The creation of image feature information base improves the working efficiency of the system and saves the cost of human and financial resources.