Color image segmentation—an innovative approach

Abstract

In this paper we describe a color image segmentation system that performs color clustering in a color space and then color region segmentation in the image domain. For color segmentation, we developed a fuzzy clustering algorithm that iteratively generates color clusters using a uniquely defined fuzzy membership function and an objective function for clustering optimization. The fuzzy membership function represents belief value of a color belonging to a color cluster and the mutual interference of neighboring clusters. The region segmentation algorithm merges clusters in the image domain based on color similarity and spatial adjacency. We developed three different methods for merging regions in the image domain. Unlike many existing clustering algorithms, the image segmentation system does not require the knowledge about the number of the color clusters to be generated at each stage and the resolution of the color regions can be controlled by one single parameter, the radius of a cluster. The color image segmentation system has been implemented and tested on a variety of color images including satellite images, car and face images. The experiment results are presented and the performance of each algorithm in the segmentation system is analyzed. The system has shown to be both effective and efficient.

Introduction

Advances in cognitive psychology over the past decades have revealed that visual data, in the form of scenes and pictures, are often mentally processed in visual terms alone, without any corresponding translation or recording into verbal labels or representations [1], and humans often respond strongly to color cues within image contents [2], [3], [4], [5], [6]. In the past decade, color imaging and printing devices has become more affordable and computer power has been ever increasing. As a result, color imaging has become popular in many applications including object classification and recognition, video surveillance, image indexing and retrieval in image databases, feature based video compression, etc. [7], [8]. In some applications, image contents can be better described in terms of color features combined with spatial relations such as enclosure and adjacency due to the irregularity of object shapes in images. This paper describes our research in color image segmentation, which is often a necessary computational process for color-based image retrieval and object recognition.

Image segmentation is a process of partitioning image pixels based on selected image features. The pixels that belong to the same region must be spatially connected and have the similar image features. If the selected segmentation feature is color, an image segmentation process would separate pixels that have distinct color feature into different regions, and, simultaneously, group pixels that are spatially connected and have the similar color into the same region. In color imagery, image pixels can be represented in a number of different color spaces e.g. RGB, XYZ or LUV [5], [9], [10]. One major concern in color image segmentation is that the computational complexity, which has increased significantly in comparison with gray scale image segmentation. The process of image segmentation can be considered as an unsupervised clustering, if a priori knowledge about the number and type of regions present in the image is not available [11], [12]. Image clustering procedures often use the individual image pixels as units and compare each pixel value with every other neighboring pixel value, which require excessively long computation times for images of high resolutions. A high quality representation of a color requires 8 bits using pulse code modulation (PCM) quantization for each of the three color components, Red (R), Green (G), Blue (B). For each image pixel, 24 bits of amplitude quantization is required, allowing specification of 2²⁴=16,777,216 distinguishable colors. This leads to high computational cost in image clustering.

In this paper we describe a color image segmentation system that performs color clustering in a color space and then color region segmentation in the image domain. The paper is organized as follows. Section 2 will give an overview of the color image segmentation system. Section 3 will describe a fuzzy clustering algorithm that iteratively generates color clusters using a uniquely defined fuzzy membership function and an objective function for clustering optimization. The fuzzy membership function represents belief value of a color belonging to a color cluster and the mutual interference of neighboring clusters. Section 4 will present a region segmentation algorithm that groups clusters in the image domain based on color similarity and spatial adjacency. We developed three different methods for merging regions in the image domain. Unlike many existing clustering algorithms, the image segmentation system does not require the knowledge about the number of the color clusters to be generated at each stage and the resolution of the color regions can be controlled by one single parameter, the radius of a cluster. The color image segmentation system has been implemented and tested on a variety of color images including satellite images, car and face images. Section 5 presents our experiment results and the performance analysis of each algorithm in the segmentation system. The system has shown to be both effective and efficient.