A comparative study of image low level feature extraction algorithms

Feature extraction and matching is at the base of many computer vision problems, such as object recognition or structure from motion. Current methods for assessing the performance of popular image matching algorithms are presented and rely on costly descriptors for detection and matching. Specifically, the method assesses the type of images under which each of the algorithms reviewed herein perform to its maximum or highest efficiency. The efficiency is measured in terms of the number of matches founds by the algorithm and the number of type I and type II errors encountered when the algorithm is tested against a specific pair of images. Current comparative studies asses the performance of the algorithms based on the results obtained in different criteria such as speed, sensitivity, occlusion, and others. This study addresses the limitations of the existing comparative tools and delivers a generalized criterion to determine beforehand the level of efficiency expected from a matching algorithm given the type of images evaluated. The algorithms and the respective images used within this work are divided into two groups: feature-based and texture-based. And from this broad classification only three of the most widely used algorithms are assessed: color histogram, FAST (Features from Accelerated Segment Test), SIFT (Scale Invariant Feature Transform), PCA-SIFT (Principal Component Analysis-SIFT), F-SIFT (fast-SIFT) and SURF (speeded up robust features). The performance of the Fast-SIFT (F-SIFT) feature detection methods are compared for scale changes, rotation, blur, illumination changes and affine transformations. All the experiments use repeatability measurement and the number of correct matches for the evaluation measurements. SIFT presents its stability in most situations although its slow. F-SIFT is the fastest one with good performance as the same as SURF, SIFT, PCA-SIFT show its advantages in rotation and illumination changes.