This project focuses on evaluating and comparing various feature extraction methods—SIFT, ORB, and KAZE—for object recognition within a dataset of images.
The process involves loading and splitting the dataset into training and testing sets, applying each feature extraction method to detect keypoints and compute descriptors, and then using these descriptors for object recognition via the k-Nearest Neighbors (k-NN) matcher.
The performance of each method is assessed and visualized through confusion matrices, providing insights into their accuracy and effectiveness in correctly identifying objects.