Relative shape context based on multiscale edge features for disaster remote sensing image registration
Shumei Zhang, Jie Jiang, Shixiang Cao · 2012
When disasters occur, image local gradients change significantly, whereas global shapes and structures remain relatively stable. Considering the low matching positive ratio in original SIFT, this paper propose a novel registration algorithm using Relative Shape Context (RSC) based on multiscale edge features. Firstly, edge features of global shapes and structures are extracted. Then an equivalent difference of Gaussian (DOG) space is used to detect local scale invariant features of multiscale edge images. Finally, RSC is performed as feature descriptor to find matching points. Experimental results show that the new algorithm is suitable for multiscale images and is invariant to a range of rotation angle changes. The distinctive novel algorithm owns much higher matching accuracy and is more stable than original SIFT.