Image object classification using saccadic search, spatio-temporal patternencoding and self-organisation

A method for extracting features from photographic images is investigated. The input image is through a saccadic search algorithm divided into a set o f sub-images, segmented and coded by a spatio-temporal encoding engine. The input image is thus represented by a set of characteristic pattern signatu res, well suited for classification by an unsupervised neural network. A st rategy using multiple self-organising feature maps (SOM) in a hierarchical manner is used. With this approach, using a certain degree of user selectio n, a database of sub-images is grouped according to similarities in signatu re space. (C) 2000 Elsevier Science B.V. All rights reserved.