Pattern recognition using evolution algorithms with fast simulated annealing

Abstract

This paper proposes a hybrid pattern recognition system based on evolution algorithms with fast simulated annealing that can even recognize patterns deformed by the transformation caused by rotation, scaling, or translation, singly or in combination. The proposed method rests on a polygonal approximation technique, which extract appropriate feature vectors of specified dimensions characterizing a given shape. The features are utilized as inputs in a matrix's based classifier for the shape recognition. Object recognition is formulated as matching a global model graph with an input scene graph representing either a single object or several overlapping objects. A matrix is defined as matching states for the model graph and scene graph and considered the matrix as organism of evolution algorithms, where the elements of the matrix are the possible matches. The proposed algorithms start with an initial temperature and select a finite number of organisms as parents, then employ evolution algorithms to produce a new population of organisms according to evolution rules, such as selection, mutation. After those sequential generation of organisms are produced, each of which tends to produce a superior population. By repeating above procedure for a finite number, the optimal matches from organism's population can be obtained. The proposed algorithms are aimed to solve two types of problems in the evolution algorithms. First, population in the evolution can be reduced, and second, hill climbing is facilitated, thus local minimum can be escaped. In this proposed evolution fast simulated annealing (EFSA).algorithms, the Boltzmann distribution function is replaced with the Cauchy distribution in the annealing process, so the EFSA cannot only increase the annealing speed more quickly at about [1+ t]/[1+log (1+ t)]but also have more opportunity to escape from the local. The above reason is that the wing of the Cauchy probability function has reached the deeper valley, the Boltzmann probability function has negligible value and thus has less change to escape from the local minimum.