Computer analysis of visual image similarity

Abstract

This paper is a description of image analysis and machine-learning algorithms used for multiclass image classification in the process of our partici-pation in the ImageCLEF 2012 competition. Our goal was to develop an appli-cation that could successfully determine the location of a mobile robot using the visual information provided by a camera placed on the robot. The resulting ap-plication uses machine-learning methods that improved on self-organizing Ko-honen maps and classification algorithms based on probabilistic models. The result of our work was an application that was able to correctly classify 86 per-cent of input images presented in the ImageCLEF Robot Vision task.