Directional feature detection and correspondence

Abstract

A method is proposed to detect useful directional feature points other than corner points considering that the number of corner points may not be sufficient in a scene. This is achieved by directional analysis of properties of image points by virtue of the proposed gradient operators with different direction topologies. A matching criterion is also proposed to find the initial correspondence by using the feature vectors that are acquired from the results of directional analysis. For the purpose of improving the final correspondence, four constraints are employed in the system to seek and refine the correspondence.