Matrix Fuzzy-Probabilistic Neural Network in Image Recognition Task

The fuzzy matrix probabilistic neural network (FMPNN) designed to solving image classification tasks where images are represented in matrix form is proposed. This neuro-fuzzy system has four layers of data processing which are fed into the system in data stream form on the "sliding window". It is supposed that formed classes of data-images can be arbitrarily overlapped in feature space creating a situation of fuzziness. The tuning of FMPNN is the combination of learning procedures such as "Lazy learning", "Winner takes all", "Learning vector quantization", "Fuzzy C-means clustering". The feature of the proposed fuzzy classification - image recognition system is the high speed of data processing allowing to process data within the concept of Data Stream Mining.