Image thresholding using a novel estimation method in generalized Gaussian distribution mixture modeling

The generalized Gaussian distribution (GGD) mixture model is a parametric statistical model, which is frequently employed to characterize the statistical behavior of a process signal in industry. This paper considers the GGD mixture model to approximate the empirical distributions, especially for those arising from non-Gaussian sources. A new estimation method is developed for fitting the GGD mixture model. The proposed method integrates Particle Swarm Optimization (PSO) from Computational Intelligence and Entropy Matching Estimator (EME) from Statistical Computation to seek the optimal parameter estimates, particularly when there is at least one large shape parameter in the GGD mixture model. Thus, the method is termed PSO+EME. Applications to multi-level thresholding in image processing are used to illustrate PSO+EME. Image thresholding is a useful technique to separate the interested object from background information. Due to the versatility of the GGD mixture model in characterizing process signals, it is chosen to fit the intensity of image and PSO+EME is used to estimate the parameters. The experimental study shows that the fitted model produced by PSO+EME could depicts quite successfully the non-Gaussian probability density function of image intensity, and therefore present quality effectiveness in the problem of multi-level thresholding.