Image registration based on SOFM neural network clustering

Spacecraft autonomous navigation method based on remote sensing image is a novel method which has been put into practical use in recent year. However, with the restricted computation resource onboard, this method still needs to look for a better way to improve its speed to satisfy real-time requirement. According to the problem above, this paper proposed an alternate method to faster image registration progress with the combination of SURF (Speed-Up Robust Features) and SOFM (Self-Organizing Feature Map). SURF is a sophisticated algorithm which is invariant to scaling and rotation, and SOFM is a kind of neural network which is used for clustering analysis in this paper. Then, a new algorithm called Similarity-Clustering Algorithm (SCA) is presented to analysis the result of clustering. Besides, a performance index called Precision is defined to determine the best value of Epoch for SOFM neural network training. The experimental results demonstrate that the proposed approach for registration has good adaptability for real-time practical application.