Object recognition from multiple sensory data by neural feature extraction and fuzzy structural description

This paper applies the technique of artificial intelligence to the problem of object recognition by part decomposition and feature combination from multiple sensor data. The method is based upon structural description of objects by fuzzy rules, and biologically inspired state dependent modulation of feature extractors. Fuzzy rules are applied in the generation of state dependent modulation signals that adjust and facilitate the extraction process by scheduling the execution priority between the different feature extractors, sa well as in the combination of features obtained from multiple sources. In addition, feed-forward neural network with one hidden layer is used to extract features from image data before the process of fuzzy combination. An application of human face recognition is studied to illustrate the usefulness of the proposed methodology