Feature selection on Persian fonts: A comparative analysis on GAA, GESA and GA
Tahereh Pourhabibi, Maryam Bahojb Imani, Saman Haratizadeh · Procedia Computer Science · 2011
In This paper we compare two new optimization techniques combining Genetic Algorithm and Simulated Annealing for feature subset selection on a platform of 5000 textures of 10 different Persian fonts to obtain an optimal or near-optimal feature subset with high accuracy in classification and speeding up the time that the algorithm will take to reach equilibrium. This is the first paper to apply Simulated Annealing based optimization techniques to the problem of feature selection, especially in Persian Font Recognition. As a result of our researches, we found that two proposed algorithm, Genetic Annealing and Guided Evolutionary Simulated Annealing, can achieve better recognition rate with more decrease in number of features and Guided Evolutionary Simulated Annealing has less convergence time comparing with classic Genetic Algorithm and Genetic Annealing, because of several parallel Simulated Annealing chains.