Scale-Adaptive Feature Points Extraction and Dual-Scale Based High-Precision Image Registration

The SIFT algorithm had a profound impact on the area of image local feature extraction and computer vision, since its inception. Because of its robustness and anti-interference performance, it has been widely applied in engineering. However, in practical usage of SIFT, there are actually no sufficient feature points that can be extracted to perform the image registration. For tiny scale targets, this situation becomes even worse. Based on the analysis of the scale-space, this paper puts forward a scale-adaptive feature points extraction method (SFPE). SFPE can extract more feature points than traditional SIFT algorithm, especially when the target scale is tiny. Furthermore, a dual-scale based high-precision image registration process (DHIR) is proposed in this paper. In DHIR, a small-scale matching (precise matching) process is performed after the large-scale registration. Experimental results show that the DHIR is more effective to eliminate mismatches than traditional SIFT algorithm.