Harris Correlation Descriptor (HCD): A Novel Descriptor for Point Matching

In this paper, a novel descriptor for point matching, called Harris correlation descriptor (HCD), is proposed. Inspired by the Harris feature detector, we use the Harris correlation measure defined with the determinant and trace of the Harris correlation matrix to characterize the gradient distribution in a neighborhood of feature points, and then construct the HCD descriptor which is invariant to image rotation and linear change of intensity. The using of the gradient mean in the Harris correlation measure makes the HCD descriptor not sensitive to the estimated main orientation of feature points, thus robust to image rotation. Moreover, the HCD descriptor has also a good adaptability to other image transformations.