Automatic region segmentation method based on estimating intensity-distribution parameters

Abstract

We developed a method for initial region segmentation based on the mixture Gaussian model. First, the distribution parameters corresponding to objects in a scene are estimated from intensity histograms by using the EM algorithm. Next, the range parameters for region segmentation are determined by using the estimated distribution parameters. Finally, that images are segmented by using these range parameters. Experimental results obtained by using real scenes from a PC program showed that the method was effective for the initial region segmentation task. This method can be used in a variety of object recognition systems applied for various industrial applications, including surveillance, in fixed ground sub-systems of intelligent transport systems (ITS), and in welfare support systems.