A.V. Vartanov, E.A. Malyasova

For the development of bionic approach to the problems of automated recognition and categorization of complex objects with varying shape it is necessary to clarify how the process of categorization occurs in humans. The paper compares two kinds of categorization of a stimuli - categorization of the object and its type of transformation. There are two main difficulties in categorization study planning: existing of early systems of categorization, which couldn-t be found out by popular non-invasive methods (fMRI and PET) because of its low temporal resolution; and in case of EEG study - large number of incentives (for the possibility to separate objects into groups), which leads to long duration of ERP experiment. So we proposed modified method of dipole localization which can solve these problems. According to the data of the EEG study, processed using the proposed method, space-time model (structure) activation of brain structures was constructed. It demonstrates the specificity of the mechanism for determining the type of transformation as compared with the task of object identification. The structures provided the general pretuning of perception and structures associated with a more detailed analysis were detected. Structures whose activity varies significantly across the range of analysis depending on the task (mainly frontal cortex), appear to be responsible for «screwing» of perception. Structures whose activity varies in a series on some (mainly in the middle) intervals are likely to be responsible for the selection of significant attributes necessary for the proper classification of the object to one category or another (the structures of the temporal lobe, left occipital gyrus). There is a hemispheric asymmetry of the process of naming the figures (the predominance of left hemisphere activity) and their transformations (right hemisphere); Differences in the categorization processes are also confirmed by the method of averaged evoked potentials. So the proposed modified method of dipole localization is adequate for the study of categorization. The results obtained by this method are consistent with the results obtained by classical methods of EEG processing (ERP), but are not limited to them.