Research On Image Local Invariant Feature Description And Matching

Image feature matching is a major subject of computer vision and pattern recognition, and it is also an important step in the other computer vision applications,such as image registration, object recognition, image splicing and 3d reconstruction, etc.,so the research on feature matching has been widely studied both in academia and industry. While in practical problems, the image scenes are different, also changes such as scale, viewpoint, rotation, translation, illumination are existed in images. Among them, image deformation and repetitive pattern are two main aspects. These factors affect the performance of feature matching seriously. So the study on improving descriptor’s robustness to various image changes is also a challenge subjects.This paper studies feature description and matching on images with distortion and repetitive patterns. It includes a local invariant feature description method and a geometric guided-constraint based matching method. The novelty lies in two aspects:(1) Based on feature points pairing, the paper proposed a local invariant feature matching method. The distance information between features is used to compute feature pair’s support region size, which is different from the methods by using detectors to provide information for finding support region. The rotation invariance is introduced by using a sub-region division method based on intensity order. Thus it avoids the procedure of main direction estimation, so it shows better performance to image rotation.For comparison to the popular descriptors SIFT and SURF, we also choose the detected points by detectors Fast-Hessian and DoG as feature points to start our method,respectively. And additional experiments are also compared to methods similar to the proposed method like Tell’s and Fan’s. The results of the experiments have shown that the proposed method giving better performance to various image changes, especially on images with scale and viewpoint transformations.(2) To resolve feature matching problem in images with repetitive patterns, the paper gives a feature matching method, called guidance-constraints method(GCM),which has obvious advantages in resolving problems of matching features on images with image distortions or repetitive patterns. In GCM, feature points are grouped intopairs and connection compatibility is introduced to describe the relative geometric relations among features, and then potential matches are found by using the defined geometric guidance and are verified by using the defined geometric constraints.Experimental evaluation shows that the proposed GCM can significantly improve both the number of correct matches and correct ratio under various image transformations, its effectiveness has been proved in applications on images with distortions or containing repetitive patterns.