Morphology-based object coding: self-similarity detection via pattern sensitive image sampling
Koji KAMEJIMA · 2002
Intended for reactive access to open information space, a method is presented of object description in noisy background. The description is transparent so that visible information is contoured and parametrized without any a priori information. For this purpose, visible images are first approximated by fractal attractors, without serious loss of generality. By evaluating capturing probability of unknown fractal attractors, the most probable points are then extracted to restore self-similarity structure to be identified. Finally, by using the proposed method a computer vision system is shown to be equipped with the capability for "picking up" not-yet-identified objects, and as a result, the robot can interact with unknown objects under a complete control of the human operator.