Research On Image Matching Algorithm Based On Point Feature

With the development of image acquisition technology,such as UAV,satellite and digital camera,people have more and more means to acquire image data,which is more and more convenient,and the quality of image data is also getting higher and higher.The content and information provided by a single image often can not meet the needs of current production applications.Therefore,in order to obtain deeper information in the target image,it is necessary to register multiple target images.In real life applications,because of the complexity of scenes and the diversity of images,traditional image registration algorithms can not meet the needs of high accuracy and real-time.In this paper,image registration technology is studied based on image features.The advantages and disadvantages of traditional algorithms are analyzed through experiments,and their shortcomings are improved.A fast and high-precision image registration algorithm and mosaic scheme are summarized.The contents of this paper are as follows:(1)In this paper,a variety of feature point detection algorithms are compared and analyzed.Harris algorithm is used to improve the ORB feature point detection method,and the feature point extraction process is optimized to simplify the image feature point extraction process.The experimental results show that the improved algorithm time is shortened by 1/2,effective Improved feature point detection efficiency.(2)In the aspect of scale attributes of feature points,a simplified scale space structure is constructed.The coordinates and scale of the feature points are extracted by interpolation fitting,which provides stable scale attributes for the image feature points.The multi-scale detection of feature points is realized through experiments.(3)The gradient direction histogram method is used to replace the gray center centroid method to solve the main direction of the feature points,and the feature image window of the main direction is improved to improve the accuracy of the main direction of the feature points.(4)The binary feature description mode of the ORB algorithm is improved,and the feature points are characterized by the scale attribute information of the feature points.Experimental results show that the improved ORB feature descriptor has stable scale descriptiveness.(5)The block random sampling is used to optimize the RANSAC algorithm,and the double loop criterion is set to increase the cycle upper limit,reduce the number of iterations,and reduce the time complexity of the algorithm running.The experiment proves that the improved algorithm reduces the speed by 896.4ms under the premise of ensuring the same calculation accuracy,which indirectly improves the accuracy of the homography matrix calculation.This paper uses a large number of different types of image data to test and demonstrate the improved algorithm.The experimental results show that the correct matching rate of the matching algorithm based on the ORB scale and the main direction is improved by 55.41%,and the correct rate of rotation registration is increased by 37.73%.The improved RANSAC algorithm based on block detection reduces the time consumption by 896.4ms.Both algorithms can effectively improve image registration efficiency and provide a fast and efficient combination algorithm for image registration and image stitching.