A modified automatic registration method for large mismatching images based on MSERs features

作者：Yao Guobiao; Deng Kazhong^*; Yang Huachao; Ai Haibin; Du Quanye

来源：Journal of China University of Mining & Technology, 2014, 43(3): 540-546.

• Automatic registration
• Feature regularization
• High-precision
• Maximally Stable Extremal Regions
• Projective transformation
• Self-adaptive
• Statistical characteristics
• Sub-pixel registrations

A modified automatic sub-pixel registration algorithm is proposed based on Maximally Stable Extremal Regions (MSERs). Firstly, the MSERs which would be fitted by ellipses were detected on gray image sequence, then the dominant gradient orientation of feature neighborhood was replaced by epipolar ray orientation, and the improved method of feature regularization and description was proposed in this paper. Considering the distribution of matching feature, a new self-adaptive feature matching strategy was implemented, and thus the reliable matching set was obtained. Secondly, the least square matching method was adopted to produce fine matching points, and the mismatches were rejected based on deviation statistical characteristics and the criterion of three times standard deviation. Finally, the projective transformation model was established based on the precise correspondent points. The results have confirmed that the proposed algorithm is superior to the available method for images characterized by severe rotation, large scale and view change.

• 出版日期2014
• 单位中国矿业大学（北京）