Particle swarm optimisation method for texture image retrieval

Abstract

There are two important tasks in texture image retrieval systems namely, feature extraction and similarity measurements. Two essential requirements of texture image retrieval system are immense retrieval precision and reduced computational complication. Several efficient methods for texture feature extraction and similarity measure methods exist. Objective of the present chapter is to propose efficient texture feature extraction algorithms which should have high retrieval accuracy. Different orthogonal moment can represent an image with almost zero information redundancy. Calculation complexity and approximation error with Zernike moment is very high. So in our work to extract feature Exact Legendre Moment (ELM) has been used. In the present chapter a new search algorithm using particle swarm optimisation (PSO) has been presented. Out of global best and local best model of PSO global best model has been considered here. The proposed method has been compared with energy, standard deviation and energy + standard deviation based retrieval method. To improve the performance of search method different modifications have been proposed. Four texture image searching algorithms have been provided using four of these modifications namely adaptive inertia weight PSO, guaranteed PSO, improved PSO and attractive repulsive PSO in this chapter. These modified methods have been compared with some existing retrieval methods like M-band wavelet, Cosine modulated Wavelet based texture image retrieval system.