An Adaptive Approach for Keypoints Description Using Fractional Derivative

Traditional image descriptors tend to utilize integral derivative characterizing local features, like orientation histogram. However, integral-based derivative has a disadvantage in describing image texture details in smooth area. In this paper, we propose a novel framework for reestablishing orientation histogram based on adaptive fractional derivative, which is better at representing local feature. Then a general weighting scheme for orientation histogram is developed, which improves the accuracy of keypoints description. Finally, we demonstrate the utility of our formulation in implementing solutions for various keypoints descriptions tasks. To exercise our framework we have created a new SIFT and SURF application over images.