A Comparative Study of Feature Detection and Description Algorithms for Computer Vision Applications: Assessing Accuracy and Computational Efficiency

Feature detection and description algorithms are the most widely used techniques in various computer vision works such as panorama image stitching, object tracking, and object detection. This study brings a novel method to compare algorithm’s accuracy to detect and describe features. Computational efficiency of the algorithms is also determined. Feature detection algorithms that are analyzed in this study are SIFT, BRISK, ORB, KAZE and AKAZE. In this analysis, five datasets have been considered: built-up area, mountain, forest, water body and barren land. In order to analyze the robustness of these algorithms, every second image of these datasets are distorted by creating noise and features are determined. Noises created are the visual and viewpoint changes on the image; such as radiance change, scale change and orientation change. Algorithms differ in their functionality, SIFT linearly diffuse entire image while KAZE and AKAZE performs nonlinear diffusion. Whereas BRISK and ORB algorithms differ in features description by using binary descriptors. The result obtained from the research shows that ORB is the most accurate and computationally efficient algorithm to determine the features.