Performance Characterization of Image Feature Detectors in Relation to the Scene Content Utilizing a Large Image Database

摘要

Selecting the most suitable local invariant feature detector for a particularapplication has rendered the task of evaluating feature detectors a criticalissue in vision research. Although the literature offers a variety ofcomparison works focusing on performance evaluation of image feature detectorsunder several types of image transformations, the influence of the scenecontent on the performance of local feature detectors has received littleattention so far. This paper aims to bridge this gap with a new framework fordetermining the type of scenes which maximize and minimize the performance ofdetectors in terms of repeatability rate. The results are presented for severalstate-of-the-art feature detectors that have been obtained using a large imagedatabase of 20482 images under JPEG compression, uniform light and blur changeswith 539 different scenes captured from real-world scenarios. These resultsprovide new insights into the behavior of feature detectors.