Research Of Image Correction And Matching Algorithm For Engineering Application

Image correction and matching,as a basic problem in the field of image processing,is one of the key technologies for image preprocessing.It has been widely used in many practical engineering fields such as medicine,remote sensing,military and binocular stereo vision.Image correction is the process of recording and superimposing two or more images in the same scene obtained under different conditions such as different time and sensor devices.At present,SIFT(Scale Invariant Feature Transform)feature and Random Sample Consensus(RANSAC)algorithm were frequently used in image correction.However,It consumed a lot of time to extract SIFT features and to eliminate false matches with RANSAC algorithm,and there are still a few errors after false matching culling.Image stereo matching is a classic computer vision problem.Its goal is to find the matching corresponding points from the left and right viewpoint images of the same scene captured by the binocular camera,and obtain the disparity with the stereo matching algorithm to estimate the image depth information.Due to the camera itself,the left and right viewpoint images will be optically distorted,so image correction must be performed before stereo matching.The accuracy and real-time performance of the traditional method or Convolutional Neural Network(CNN)method for stereo matching cannot meet the requirements of engineering practice online application.In response to the above questions,the main works of this paper are as follows:(1)In the image correction and recognition,the enterprise actually adopts the SIFT algorithm,the feature extraction process consumed a lot of time,the operation efficiency is low,there are still a few errors after using the RANSAC algorithm to eliminate the mismatch,resulting in unsatisfactory image correction effect.Aiming at this problem,the SURF(Speeded Up Robust Features)feature was adopted to replace the SIFT feature in this paper,which greatly improved the feature extraction efficiency and speeded up the image processing process.At the same time,the improved RANSAC algorithm was adopted.Firstly,the ratio of the nearest neighbor distance to the next nearest neighbor distance was used as the threshold to determine probably the correct match,and then the improved RANSAC culling algorithm was used to clean out the mismatch and improved the accuracy of the correction.(2)In order to further improve the operational efficiency and accuracy of imagecorrection,an image correction algorithms was proposed based on ORB(Oriented FAST and Rotated BRIEF)feature and Motion consensus.Firstly,feature points were extracted with ORB algorithm,and initial matching of feature points was implemented based on Hamming Distance.Secondly,the Motion consensus algorithm was used to eliminate false matching,and then topological constraint pair was used to eliminate stubborn mismatches.Finally,the transform matrix H was calculated using the RANSAC algorithm to complete the image correction.The experimental results show that the algorithm is not only faster in feature detection and matching,but also has better robustness than SIFT algorithm.(3)The accuracy and real-time performance of the traditional method or convolutional neural network method for stereo matching cannot meet the requirements of engineering practice online application.Therefore,a real-time adaptive stereo matching network algorithm was proposed in this paper.The accuracy and real-time performance was improved by introducing a new lightweight and effective architecture Modularly Adaptive Stereo Network(MASNet),embedded an unsupervised loss function and residual refinement module.Firstly,the multi-scale features were extracted by the pyramid network.Then the initial disparity estimation was carried out,and finally,the final disparity map was output by the residual refinement module.The experimental results show that the proposed method is more accurate than the model with similar complexity,and the processing speed of about 25 frames per second on average meets the requirements of engineering practice online usage.