Color Image Segmentation Using Gaussian Mixture Model and EM Algorithm - Multimedia and Signal Processing - page 61

Color Image Segmentation Using Gaussian Mixture
Model and EM Algorithm

Zhaoxia Fu 1,2 and Liming Wang 1

1 Science and Technology on Electronic Test &
Measurement Laboratory and Key Laboratory of Instrumentation Science &
Dynamic Measurement(Ministry of Education),
Information and communication engineering institute,
North University of China, Taiyuan, 030051, China
2 Party school of shanxi provincial committee of the C.P.C, Taiyuan, 030006, China
[e-mail], [e-mail]

Abstract. The segmentation of color image is an important research field of
image processing and pattern recognition. A color image could be considered as
the result from Gaussian mixture model (GMM) to which several Gaussian
random variables contribute. In this paper, an efficient method of image
segmentation is proposed. The method uses Gaussian mixture models to model
the original image, and transforms segmentation problem into the maximum
likelihood parameter estimation by expectation-maximization (EM) algorithm.
And using the method to classify their pixels of the image, the problem of color
image segmentation can be resolved to some extent. The experiment results
confirm this method validity.

Keywords: Gaussian mixture model, EM algorithm, image segmentation,
random variable.

1 Introduction

Image segmentation is the most important issue on automatic image analysis and
pattern recognition. According to some features of the image or similar criteria of
feature set, it is carried on the grouping cluster to the pixels, and the image is divided
into a series of meaningful and different regions. The quality of image segmentation
and regional boundaries precision immediately influence the following region
description and image analysis and understanding, and they are important technical
aspects in the image analysis, processing, understanding.
Traditional techniques of image segmentation can be divided into edge-based
segmentation and region-based segmentation. The former is called for edge extraction
from the target segmentation objects. The latter often determines the object boundary
through spatial partial characteristic of the image, such as gray, texture, and other
statistical properties of pixels. EM algorithm is a region segmentation method based
on statistical pattern recognition, which has speediness and wide adaptability. The
paper proposes a method of image segmentation based on Gaussian mixture model