New Feature Correspondence Method using Bayesian Graph Matching Algorithm

In this paper, we propose a new approach for establishing the correspondence between local invariant features using Bayesian graph matching algorithm. First, we will discuss various local invariant feature detectors and descriptors for scale and affine transformation and illumination changes. Second, we propose an efficient features corresponding method using local invariant features and new graph matching algorithm. Finally, we evaluate the comparative performances of our method (BGMA) with several existing feature matching approach such as Lowe’s shift matching by conducting experiments on various real image data. We test on three image pairs taken from MSRC v2 dataset and Caltech 101 dataset. Experimental results show that the proposed method clearly outperforms rather than the existing matching algorithms about feature correspondence in images with rotation or scale transformation and illumination changes.