Simulating preattentive and attentive vision with Moore-Penrose associative memories

The Moore-Penrose model of distributed associative memory (DAM) has been described previously as a powerful method for pattern recognition. It is shown that it also can be used for preattentive and attentive vision, and provides a mathematical analysis of the properties leading to this application. The basis for the preattentive system is that both the visual input features as well as the memory are arranged in a pyramid. This enables the system to provide fast preselection of regions of visual interest. The selected areas of interest are used in an attentive recognition. The reason for application of the DAM is based on a statistical theory of rejection. The availability of a reject option in the DAM is the prerequisite for novelty detection and preattentive selection. It can be used both in a supervised and an unsupervised learning system. Experimental results prove the feasibility and benefits of the improved recognition method