Neural-like network model for color images analysis systems

In this paper we present the chromatic neural-like network. It is a two-layer network architecture used in an image analysis system to learn objects classification tasks. Each processing unit in the hidden layer is considered as a network which codifies the color information at pixel level. The output layer makes a features analysis from the hidden layer responses in every window of the image. The architecture and operation of this network are extracted from the studies of biologic visual neural systems behavior and it is a model to be used in an artificial vision system. The learning methodology is an unsupervised, fuzzy and adaptive one and lets a faster training for image processing than other algorithms. It combines the thresholding in features spaces technique and the fuzzy Kohonen clustering nets approach with a gaussian membership function. Experimental results of the net performance with real images are shown