Rough Sets: Theoretical Aspects of Reasoning about Data
Rough Sets: Theoretical Aspects of Reasoning about Data
IEEE Transactions on Neural Networks
| Hi-index | 0.00 |
The visual systems of humans and primates outperform the best artificial vision systems by almost any measure. Humans can easily recognize as complex objects as faces even if they haven't seen them in such conditions before. However, experiments with the inverted faces (Thatcher illusion) show strong dependences between parts and their configuration. We propose pattern recognition rules similar to the primate visual brain on the basis of the simple shape classification in the intermediate area of the visual cortex (V4). In the present work we have described interactions between parts and their configurations using single cell responses as the brain expertise (decision attribute). Experimental data as the set of condition (stimulus) and decision (cell responses) attributes were placed into a decision table. Applying the rough set theory (Pawlak, 1992) we have divided our stimuli into equivalent classes determined by evoked cell responses. On the basis of the decision table, we found the decision rules. Comparing decision rules for responses to object and to its parts, we have found the interaction rules in the receptive fields of the area V4. We have proposed the interaction rules for objects that are simpler than faces but we expect that such rules can give us neurophysiological basis for the Gestalt perception of the complex objects. By comparing responses of different cells we have found equivalent concept classes. However, many different cells show inconsistency between their decision rules, which may suggest that the brain uses several different decision logics in order to make object perception insensitive to changes in properties of its parts (rough parts).