Fast Spatial Verification with Affine Transformations

A fast spatial verification scheme using affine transformations is proposed to improve the retrieval precision by exploiting proper spatial constraints among visual key words.Firstly,affine covariant neighborhoods of visual words are used to verify the spatial consistency.Secondly,a set of affine transformation matrices between the corresponding covariant regions is calculated,and the matrix with the most inliers is chosen as the best matching affine transformation matrix(BMATM). In this step,only a single pair of corresponding affine covariant regions is used to calculate the 6 degree of freedom(DOF)affine transformation for acceleration.In the last step,the spatial consistency is verified by the BMATM.Experimental results show that this scheme is more robust,more accurate and faster in compare with the state-of-the-art approaches.