Keypoint Matching Using Salient Regions and GMM in Images with Weak Textures and Repetitive Patterns
Zhu, Qidan,Wang, Feng,Cai, Chengtao,Meng, Haiyang,Qiao, Renjie
DOI: https://doi.org/10.1007/s11042-022-12503-0
IF: 2.577
2022-01-01
Multimedia Tools and Applications
Abstract:In the keypoint matching task, most standard methods cannot extract uniform and effective keypoints in images with weak textures and repetitive patterns, nor can they design unique feature descriptors for them. In order to bypass the difficulties encountered in matching ambiguous images, we make some efforts from keypoint extraction and keypoint matching respectively. First, a keypoint extractor based on local energy maxima is designed, which can extract stable features in weak texture regions. Second, to guide keypoints matching based on the Gaussian mixture model framework, an adaptive search method to locate salient regions in images without training is introduced. This is accomplished by quantifying the similarity between the center point and the remaining keypoints. Finally, the iterative equations considering the guiding information is derived. These algorithms alleviate the impacts of weak texture and repetitive patterns on keypoints matching. Experiments on test data show that our method can extract stable and reproducible keypoints, in matching, the precision is more than 60% and the recall is more than 80% when images containing description ambiguities.