A local image descriptor based on radial and angular gradient intensity histogram for blurred image matching

Abstract

Image rotation and scale change can significantly degrade the efficiency of local descriptors in blurred image matching. Conventional local image descriptors often only employ the rectangular gradient information of detected region around each interest point. Due to unwanted errors estimated for scale and dominant orientation, the performance of these local descriptors is severely degraded when applied to blurred images. To solve this problem, we propose a novel descriptor called radial and angular gradient intensity histogram (RAGIH) which jointly utilizes gradient and intensity features. In this local descriptor, feature vectors are extracted from two concentric circular regions around each key point and using angular and radial gradients in a specific local coordinate system reduces the estimation errors. Extensive experiments on challenging Oxford dataset demonstrate the favorable performance of our descriptor compared to state-of-the-art approaches. © 2018, Springer-Verlag GmbH Germany, part of Springer Nature.