Event Driven Motion-Image Classification by Selective Attention Model
Toshikazu Wada, Takekazu Kato · 1996
A motion-image classification method is presented. Our method is designed based on selective attention model which dynamically changes its focusing regions (domains of feature extraction) according to its state so that the essential features can be extracted from input images. The advantages of our method are 1) the feature extraction is not affected by the image variations outside of the focusing regions, 2) feature extraction and the state transition can he complited much faster than other methods, 3) focusing-region sequence can be learned incrementally from training samples, 4) the classifier can he derived from notion-image identifiers trained independently, and 5) the classifier does not output unique result but possible candidates when the input is ambiguous.