Аннотация:The method to describe in symbolical way the images as structural spectra of image’s labeled
graph. Each graph’s vertex corresponds to local singularity found in the image is proposed. Additional markers are assigned to each graph’s vertex on the base of the local image singularity classification. The algorithm to build “image-features” matrix X is presented where feature column described the repetition number of some graph’s fragment. Cases of connected and disconnected fragments are considered. Using the processing chain: (1) “image features” matrix X building (2) construction of the classification function (of given type) (3) validation of image classification, one can in an iteration manner select the image description parameters which will be adequate for the given image classification task.