ART2 Network Based Color Pixel Categorization and Its Application

Pixel analysis is the primary step for other relevant processes of image processing. So how to distinguish pixels with different color and luminance in an image or a video stream and how to categorize them correctly and effectively are of the most challenging aspect in these field. With a reasonable mapping operation to vectors in RGB color space, this paper proposes an layered detecting approach to categorize spatial or temporal pixels series by improving ART2 network. The unsupervised self-organizing ability of ART2 makes the categorizing process more intelligent and effective. The processing steps and final experiment results illustrate that the bionic processing of ART2 network not only coheres with human psychological and physiological process of observing an image, but also has a strong adaptability for shadow, ghost and clutter noise suppressions