Attraction based recognition

An approach to image recognition is proposed, which is achieved through the correspondences between predicted and measured properties: points, lines, regions, color, shape, etc. Image parts are subjected to a series of transformations performed by the recognition process. In these transformations, called image chaining, there is a grouping of image parts or features due to a field continually perturbed by the recognition process. A fundamental role is played by the spatial correspondences together with an intentional behavior of the transformed visual data called attraction. After a definition of the attraction principle, the paper describes the attraction process, and justifies the image chaining concept. Finally, a series of processes - segmentation refinement, search space reduction, and perceptual organization - are interpreted in terms of the attraction principle.