Robust visual correspondence computation using monogenic curvature phase based mutual information

Visual correspondence has been a major research topic in the fields of image registration, 3D reconstruction, and object tracking for some decades. However, due to the radiometric variations of images, conventional approaches fail to produce robust matching results. The traditional method of intensity-based mutual information performs very good for global variations between images, however, its performance degrades in the case of local radiometric variations. Monogenic curvature phase information, as an important local feature of the image, has the advantage of being robust against brightness variation. Hence, in this Letter, we propose an approach to compute the visual correspondence by coupling the advantages of mutual information and monogenic curvature phase. Experimental results demonstrate that the proposed approach can work robustly under radiometric variations.