Cascade Windows in Intellectual Agents of Multichannel Images Classification

摘要

The paper suggest using not slid cascade windows generating "weak" qualifier for the classification of hardly structured images. At each level of processing of a parental cascade window it is divided into several window descendants, for analysis of which the intellectual agents are used. The agents use two approaches to form descriptors, creating two groups of qualifiers based on these descriptors in a cascade window and the subsequent aggregation of the decision made by these qualifiers. It is shown that the use of cascade windows in the analysis and classification of multichannel images or their segments as integration is the most efficient approach. In this case, is possible in the narrow spatial range of a cascade window. In those channels, in which the image has good spatial resolution, the descriptors are formed based on the analysis of border contours of segments of the corresponding raster. For the analysis and classification of the allocated contours rasters are used in the channels with the good permission in the field of spatial frequencies or in the field of an electromagnetic range. The use of cascade windows in each channel allows forming a set of qualifiers on one channel in the form of trees of decisions with the subsequent aggregation of decisions both in the channel, and between channels. As a result, the network structure of qualifiers (cellular qualifiers) is formed. Its parameters are defined by means of training on the basis of expert estimates or hybrid methods.