The Impact of Parameters on the Efficiency of Keypoints Detection and Description

This paper comprehensively investigates the efficiency and performance of the keypoints detection and description methods in computer vision and image processing. Four widely used methods - SIFT, SURF, ORB, and BRISK are compared and analyzed. The study aims to assess the influence of different parameters on detecting and describing keypoints, specifically focusing on the limitation of the number of keypoints and Lowe's ratio values. The effectiveness of each method is evaluated by analyzing the similarity coefficient for pairs of images with varying degrees of similarity. It has been found that the similarity coefficient is also rising, with Lowe's ratio rising. At the same time, both Lowe's ratio and detected keypoints number limitation significantly impact the achieved methods' performance. The results obtained from the experiments provide valuable insights into the pros and cons of each keypoints method. Furthermore, we suggest how to optimize parameter settings to achieve optimal performance. The findings of this research can benefit researchers and developers working in computer vision and image processing.