doi:10.3850/978-981-08-7618-0_0351

ABSTRACT

In this paper, we develop and analyze a segmentation method using Bivariate Finite Doubly Truncated Gaussian Mixture Model with K-Means and EM algorithm. Image Segmentation is used in many applications; with Image Retrieval it is possible to analyze the images in Geo-Information Systems, Medical Image Analysis, Film and Photo Analysis etc. It is customary to consider the pixel intensity only in many of the Image Retrieval methods; however it is highly important that another character of the pixel, namely, brightness plays a dominant role in Image Retrieval. It is also to be observed that the pixel intensity and brightness are correlated and follows a Bivariate Distribution and having finite range. Hence it is needed to develop an Image Retrieval method by considering that the pixel intensities and brightness in the image region follows a doubly truncated Bivariate Gaussian Distribution and the entire image is characterized by Finite Truncated Bivariate Guassian Mixture Model. In this paper, using the K-Means algorithm, the number of image regions is identified and the model parameters inside the image regions are estimated by using the “EM+MML” algorithm. The segmentation of the pixels is carried by maximizing the component likelihood.

Keywords: Doubly Truncated Gaussian Mixture Model, MML criterion, EM Algorithm, Image Quality Metrics, Image Segmentation.

© 2011 ICCA 2010 Organizers