Robust scene matching method based on sparse representation

摘要

This article presents two efficient scene matching methods based on sparse representation, which are robust to noise and occlusion. The process of searching for the correct matching position in the reference image can be considered as finding a certain atom in a dictionary to represent the vectorized sensed image. The atoms that form the dictionary are vectorized blocks which represent all the possible matching positions. We also use two steps matching to speed up the running time, and further improve the performance to some extent. Experiments demonstrate that our scene matching methods perform better than traditional methods due to the inherent robustness to corruption of sparse representation.