An adaptive alternative for syntactic pattern recognition

Currently there are three basic types of pattern recognition, the syntactic, the statistical and the neural types. They all have advantages and disadvantages, some of which are troublesome. In statistical methods it is difficult to express structured information. Neural methods have problems to represent neural networks semantically. In syntactic methods it is lead to learn new rules. This is exactly the strong point in the method presented here, due to the inherent learning ability of adaptive automata. Our method not only solves the learning problem, but also realizes that it is a promising method since it shows many of the advantages of much more complex methods, such as self-organizing neural nets, which have adaptability. The method presented here is suitable to handle robot tracking applications where the goal is to find some object or position without needing details on information of the objects or the environment. Basic geometric patterns, like rectangles and triangles are used here to ilustrate the recognition process using adaptive automata, and also to demonstrate the simplicity and efficiency of the method. This is a first step, from which it will be able to recognize a greater number and complex forms too. Keywords— Pattern Recognition, Adaptive automata, Robotics.