Image matching of junction-type feature points based on the third-order moment

Image matching of junction-type feature points in different views is a difficult process because of the complex characteristics of such corners, including significant translation, rotation, shear and scaling differences. A novel method is proposed where, initially, feature points are extracted based on Harris algorithm. Next, the third-order moment of each candidate point is used to compute its orientation vectors. Finally, an affine-invariant representation is determined by mapping the point and its orientation vectors to an orthogonal frame, and a characteristic scale is selected using the extreme of the LOG (Laplacian of Gaussians). Evaluations of the proposed method on different types of image pairs prove its efficacy with repeatability and number of correct matches (NCMs) compared to Harris-based methods.