Performance evaluation of local features for object discovery
Date
2015
Authors
Bhikadiya, Jatin V., author
Draper, Bruce A., advisor
Beveridge, Ross J., advisor
Bates, Daniel J., committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Object recognition is one of the most challenging tasks in computer vision. A common approach in recognizing an object begins by detecting local features in image using a feature detector and describing detected features in terms of feature vectors using a feature descriptor. Many local feature detectors and feature descriptors have been proposed in literature. This work evaluates performance of two successful feature detectors and five feature descriptors on three datasets with unique characteristics. Based on the information content in a given dataset we find general trends on the performance of local features. Our findings will guild computer vision practitioners selecting between alternative local feature detector and local feature descriptor to design highly accurate recognition systems.
Description
Rights Access
Subject
feature descriptors
object recognition
bag of features
performance evaluation
feature detectors