Image Object Recognition With Improved ART2

How to make image object recognition system has characteristics shown in the developmental process of human cognition is all along a difficult and hot spot in computer vision. This paper attempts to simulate and implement the process of assimilation, accommodation, equilibrium based on the perceptual schema of Sensory-Motor Intelligent Schemas(SMIS), and then establish a fast stable image object recognition and classification system with the ability of self-organization, self-learning and memory, and make the system show different learning passion when it comes across image objects with different similarity(to adjust the balance of stability and plasticity dynamically), and realize the learning, recognition and memory of the image object under supervised or unsupervised.In the aspect of image feature extraction, a novel method based on object contour is proposed in this paper. In order to realize the translation, rotation and scale invariance of the contour-based moments really, it is modified according to the method used in region-based moments for reference. On this basis four contour-based geometric features are defined and computed, and form the Mixed Contour Feature Vector(MCFV) associate with contour-based moments, which is used as the input of the improved ART2 network. The MCFV can reflect a more comprehensive characteristics of image object as a result of combining the advantages of contour-based moments features and geometric features, and have a farther distance in feature space between different image objects, that is to say, have a better discriminative ability.In the aspect of simulating the developmental process of human cognition, the ART2 network is modified by being added the activities and adjustment mechanism of vigilance. The improved ART2 network can show different learning passion when it comes across image objects with different similarity under supervised or unsupervised. On this basis, this paper achieves the simulation and implementation of the cognitive development with improved ART2 network by compared with the description of schema theory, and finally to construct an image object recognition system based on perceptual schema and improved ART2 network.?...