Skip to Main Content
In this paper we propose a probabilistic model for the local features technique which provides a methodology to improve this approach. On the other hand, a method for compensating the color variability in images is adapted for the local feature model. Finally, an experimental study is made in order to evaluate the performance of the local features approach on challenging situations such as partially occluded images and having only one training image per user. The results of the experiments are competitive with state-of-the-art algorithms even when we have the mentioned extreme situations.