Квантування ознак у методах класифікації зображень

Abstract

The study explores methods for analyzing image keypoints using the BRISK detector and quantizing them with the k-medoids algorithm. The 500 most significant descriptors were selected for each of the five images, and the reference set of descriptors was clustered to obtain a compact feature representation. This approach reduces data redundancy and enhances the robustness of feature extraction. The resulting feature distribution vectors can be used for classification, object similarity search, and other computer vision tasks, improving the efficiency of image analysis in large datasets.