Object Tagging for Human-Robot Interaction by Recolorization using Gaussian Mixture Models

In this work, we present a method to tag objects by applying a color model learned from another source object. We learn the statistical color model of objects using Gaussian Mixture Models and Expectation-Maximization algorithm. The source model is transferred to the target object to be tagged by matching the Gaussian distribution that best describe the color struc- ture. This makes the target gain the color model of the source while main- taining its initial appearance. This algorithm can be used in Human-Robot Interaction to visually tag objects for selection, targeting or discrimination. We perform some experiments to test our proposed method.