Seeded region growing: an extensive and comparative study

Abstract

Seeded region growing (SRG) algorithm is very attractive for semantic image segmentation by involving high-level knowledge of image components in the seed selection procedure. However, the SRG algorithm also suffers from the problems of pixel sorting orders for labeling and automatic seed selection. An obvious way to improve the SRG algorithm is to provide more effective pixel labeling technique and automate the process of seed selection. To provide such a framework, we design an automatic SRG algorithm, along with a boundary-oriented parallel pixel labeling technique and an automatic seed selection method. Moreover, a seed tracking algorithm is proposed for automatic moving object extraction. The region seeds, which are located inside the temporal change mask, are selected for generating the regions of moving objects. Experimental evaluation shows good performances of our technique on a relatively large variety of images without the need of adjusting parameters.

Introduction

Automatic image segmentation is an essential process for most subsequent tasks, such as image description, recognition, retrieval and object-based image compression (Majunath et al., 2000, Kunt et al., 1987). Automatic image segmentation has also become a key point of MPEG-4 and MPEG-7 standards for realizing the object-based image coding and content-based image description and retrieval. The general image segmentation problem involves the partitioning of a given image into a number of homogeneous regions according to a given critical. Thus, image segmentation can be considered as a pixel labeling process in the sense that all pixels that belong to the same homogeneous region are assigned the same label (Haris et al., 1998). The existing automatic image segmentation techniques can be classified into five approaches, namely, thresholding techniques (Lim and Lee, 1990, Sahoo et al., 1988, Pal and Pal, 1993), boundary-based methods (Kass et al., 1987, Palmer et al., 1996), region-based methods (Haralick and Shapiro, 1985, Chang and Li, 1994, Hijjatoleslami and Kittler, 1998, Adams and Bischof, 1994), hybrid techniques (Pavlidis and Liow, 1990, Haddon and Boyce, 1990, Chu and Aggarwal, 1993), and clustering-based techniques (Pappas, 1992, Shen et al., 1998).

Seeded region growing (SRG), that is introduced by (Adams and Bischof, 1994), is robust, rapid and free of tuning parameters. These characteristics allow implementation of a very good algorithm which could be applied to large variety of images. SRG is also very attractive for semantic image segmentation by involving the high-level knowledge of image components in the seed selection procedure. However, the SRG algorithm also suffers from the problems of automatic seed generation and pixel sorting orders for labeling (Mehnert and Jackway, 1997, Fan et al., 2001a). There are several potential approaches to improving the SRG algorithm:

• Scan order optimization: The original SRG algorithm uses sequential sorted list as data structure (Adams and Bischof, 1994). All pixels are put into the sequential sorted list according to their delta value. The authors in (Mehnert and Jackway, 1997) have confirmed that a different order of processing pixels leads to different final segmentation results. They also noticed two types of order dependencies. The first type is called inherent order dependencies, while the second is called implementation order dependencies. However, the unlabeled pixels may not be adjacent to all these selected seeds especially at the beginning of SRG procedure, thus the connection characteristics among the adjacent pixels should be used in the pixel labeling procedure. The objective of image segmentation is to label the adjacent (connected each other on pixel level) similar pixels with the same symbol. Since the region boundaries are used for defining the boundaries of different image components, we propose a boundary-oriented technique to accelerate the seeded pixel labeling procedure. Our boundary-oriented pixel labeling technique can also support a parallel SRG procedure.

• Automatic seed selection: The authors in (Fan et al., 2001a) have developed an automatic edge-oriented seed generation technique to automate SRG algorithm. The color edge detection technique is first performed to obtain the simplified geometric structures of a color image. The centroids of the neighboring labeled color edges are then taken as the initial seeds for region growing. However, the edge-oriented SRG algorithm may induce oversegmentation problem because the color edges may be over-detected for the texture images and thus result in redundant seeds. There are two reasonable approaches to improving this edge-oriented SRG algorithm: one is to perform a post-procedure of similarity-based region merging after the SRG procedure, and the other is to perform an image filtering procedure before the color edge detection procedure. In this paper, we will propose an automatic edge-oriented seed generation technique via image filtering.

• Temporal seed tracking: SRG is very attractive for content-based image database applications by involving the high-level knowledge of image objects in the segmentation procedure. However, automatic semantic image segmentation is an ill-defined problem because semantic objects do not usually correspond to homogeneous regions in color or texture (Deng and Manjunath, 2001). Automatic moving object extraction via seed tracking may be one reasonable solution of this problem (Grinias and Tziritas, 2001). In this paper, we propose an interesting seed tracking technique for automatic moving object extraction.

This paper is organized as follows. A brief review of SRG technique is given in Section 2. In Section 3, we propose three automatic SRG techniques (their major steps are shown in Fig. 1). A comparative study of those three automatic SRG techniques is also given. Section 4 describes a seed tracking technique for automatic moving object extraction. Section 5 introduces our techniques for semantic-sensitive salient object detection. We conclude in Section 6.