Contents I. Introduction
Image segmentation is an important technique in the image pre-processing for extracting the interesting object from the background. The existing image segmentation techniques can be classified into the following approaches, thresholding techniques, boundary-based techniques, region-based techniques, clustering-based techniques and hybrid techniques [1]–[4]. Seeded region growing (SRG) is one of the hybrid methods proposed by Adams and Bischof in 1994 [1]. It starts with assigned seeds, and grow regions by merging a pixel into its nearest neighboring seeded region. SRG is robust to the large variety of images because the characteristics of rapid and free to tune the parameters, and the considering of local information such as regions similarity, boundaries and smoothness. However, the selection of the initial seeds much influences the segmentation results. How to assign the initial seeds is the major topic in SRG. Fan and et al. [2], [3] propose a seeds selection method to assign the pixels that are between the edge regions to be the initial seeds. Shih and Cheng [4] (ASRG) use the similarity in the local region to automatically select the initial seeds. Deng and Manjunath [5] (JSEG) apply color quantization to the image in advance and then use a smoothness measurement, J value, to determine the initial seeds. Nevertheless, the unfavorable quantization leads to the poor segmentation results since the color quantization is a random process.
