Online learning of color transformation for interactive object recognition under various lighting conditions

This paper describes an online learning method of color transformation for interactive object recognition. In order to recognize objects under various lighting conditions, the system estimates a color transformation from the color of an object model by observing the color of a reference object. The system first initializes a general color transformation. Next, the system automatically recognizes a target object with the color transformation. When the system fails in recognition, the system recovers the failure with user interaction. Then the system improves the color transformation with an observed color pair of the recognized target object. By repeating, this process, the system adapts to the new environment. Experiments using real-world refrigerator scenes are shown.