Attraction based recognition

We introduce, in a model driven recognition approach, the principle of attraction which influences all transformations that may be done, departing from the first image, in order to reach the recognition of objects in a scene. The representation of knowledge has to be reduced to a representation in the feature space. This paper attempts to consider the behaviour of image transformation, for recognition purpose, in this vector space. After the definition of the attraction principle, we describe the attraction process, justify the image chaining concept, and finally interpret a series of processes, including the segmentation refinement, search space reduction and perceptual organization, in terms of the attraction principle.<
>