Research On Image Matching Method Based On ORB Algorithm

Recently, image matching has been widely applied to various fields as an important technical support, such as remote sensing image matching, moving target detection, tracking, image stitching and text recognition. Therefore, the study on precise and fast image matching has important research significance.With the development of image matching technology, the design of feature-based image matching algorithms with good invariance and robustness is becoming the research hot spot. Meanwhile, with the increasing demands on accuracy, real-time performance and the storage space, some new image matching algorithms have been proposed continuously, and the ORB (oriented fast and rotated brief) algorithm is a typical one among them. Particularly, besides the preservation of the properties of the local invariant feature-based matching algorithm, the ORB algorithm also improves the matching rate. However, as a local feature-based matching algorithm, the descriptor of the ORB algorithm has some limitations due to the local neighborhood information description of feature points. Under the influence of illumination, rotation and other interference factors, the mismatch is likely to happen due to the close change of the neighborhood of feature points. To overcome the above disadvantages, two solutions are proposed in this dissertation:the improved ORB algorithm (SFPDO) based on spatial feature point distribution descriptor and the improved ORB algorithm (RGSDO) based on the relative global shape descriptor. The specific work is as follows:1. To solve the problem of mismatch, an improved ORB algorithm (SFPDO) based on spatial feature point distribution descriptor is proposed. Particularly, the SFPDO algorithm introduce the distribution of space feature point into the feature point descriptor by using the principle that the distribution of feature points in the same area of two images is similar. The algorithm combines the spatial feature point descriptor and feature point descriptor of ORB algorithm when matching the feature points. The number of mismatching feature points decreases by weakening the limitation of the descriptor, and improves the accuracy of feature points matching.2. Through studying the effect of image edge shape in image matching, an improved ORB algorithm (RGSDO) based on the relative global shape descriptor is proposed to solve the problem of mismatch. This algorithm combines the Canny edge detection algorithm and the ORB algorithm. Firstly, the number of edge points in the circular neighborhood of the feature points is detected using Canny edge detection algorithm. In addition, the proportion of the edge points in the circular neighborhood of the feature points is calculated. Then, the proportion value is used to measure the similarity of feature points. Namely, the relative global shape descriptor in the neighborhood of feature points is combined with the feature point descriptor of ORB algorithm, which improves the matching accuracy of feature points.