KID: Kirsch Directional Features Based Image Descriptor

In these days we have seen the development of local image descriptors for several computer vision applications in order to perform reliable matching and recognition. In this direction, we have made an attempt to propose a new local descriptor which uses the Kirsch’s four directional edge features to describe the neighbourhood of the interest point. The descriptor computation mainly consists of two stages: feature detection (identification of interest points) and feature description. In the first stage, the interest points are detected using Features from Accelerated Segment Test (FAST) algorithm where interest points are identified by comparing the pixels on a circle of fixed radius around the interest point. In the second stage, the directional features for horizontal, vertical, right-diagonal and left-diagonal directions are extracted from the local region around the interest point using Kirsch masks. In order to achieve rotation invariance, the descriptor window coordinates are rotated with respect to the dominant orientation of the interest point. Experiments have been conducted on several image datasets to reveal the suitability of the proposed approach for various image processing applications. A comparative analysis with the other well known descriptors such as SIFT, SURF and ORB is also provided to exhibit the performance of the proposed local image descriptor.