Learning Variability of Image Feature Appearance Using Statistical Methods

Motivated by the problems of vision-based mobile robot map building and localization, in this work, we show that using statistical learning methods the performance of the standard descriptor based methodology for matching image features in a wide base line can be improved. First, we propose two kinds of descriptors for image features and two statistical learning methods. Later, we present a study of the performance of descriptors with and without the statistical learning methods. This work does not pretend to present an exhaustive description of the mentioned methods but to give a good idea the effectiveness of using statistical learning methods together with descriptors for matching image features in a wide base line.