Efficient and effective object recognition from a multimedia data are very complex. Automatic object segmentation is usually very hard for natural images; interactive schemes with a few simple markers provide feasible solutions. In this chapter, we propose topological model based region merging. In this work, we will focus on topological models like, Relative Neighbourhood Graph (RNG) and Gabriel graph (GG), etc. From the Initial segmented image, we constructed a neighbourhood graph represented different regions as the node of graph and weight of the edges are the value of dissimilarity measures function for their colour histogram vectors. A method of similarity based region merging mechanism (supervised and unsupervised) is proposed to guide the merging process with the help of markers. The region merging process is adaptive to the image content and it does not need to set the similarity threshold in advance. To the validation of proposed method extensive experiments are performed and the result shows that the proposed method extracts the object contour from the complex background.
Top1. Introduction
Object extraction from images and videos are an important area of research for many multimedia applications such as image and video editing, image or video retrieval and manipulation, human computer interaction and video surveillance etc. Generally, the image segmentations (Felzenszwalb, 2004; Vincent & Soille, 1991; Haris et al., 1998; Nock & Nielsen, 2004; Shi & Malik, 2000) are dividing the image pixels into the group of homogeneous small regions such that each region has visual effects and object segmentation is dividing the image into foreground and background. Image segmentation depends on low level features like color, texture and shape. Using these low-level features, we proceed for high level like object detection from image or video. Moving object detection and medical image segmentation (Li et al., 2015) are very important in daily life due to robustness and accuracy. Initially images are partitions using some low-level image segmentation into a number of non-overlapping regions. Different approaches have been used for image and medical image segmentation, Partial Differential Equation (PDE) based Active Contour Model (ACM) (Caselles, Kimmel, & Sapiro, 1997) is one important method for curve evolution. There are two main ACM based segmentation Mumford-Shah (MS) model (Mumford & Shah, 1989) and Chan-Vese (CV) (Chan & Vese, 2001) model. In last few decades, different image segmentation approaches are generally exploited to obtain object from complex scene images (Yang et al., 2008; Vanhamel, Pratikakis, & Sahli, 2003; Tab, Naghdy, & Mertins, 2006; Cour, Benezit, & Shi, 2005). The literature of images segmentation contains threshold based (Otsu, 1975) edge based segmentation (Bao, Zhang & Wu, 2005), region based segmentation (Gonzalez, Woods, & Eddins, 2004; Cheng, 1995; Peng, Zhang, & Zhang, 2011) including region splitting, region growing and region merging. These segmentations are known as low level but generally they are used as base of some high-level segmentation.
Other segmentation techniques are based on nonlinear distance metric (Sobieranski, Comunello, & Wangenheim, 2011), watershed (Haris, Efstratiadis, & Katsaggelos, 1998; Hernandez & Barner, 2000; Bleau, & Leon, 2000), homogram thresholding (Cheng, Jiang, & Wang, 2002), etc. In this context requirement of object extraction, most favourable and high level image segmentation and promising results should be obtained by less segmented regions that preserve well the boundary of objects. Some low-level image segmentation also suffers from over segmentation and under segmentation problems. In recent years, some important and interactive semi-supervised image segmentation methods are developed (Peng, Zhang, Zhang, & Yang, 2011; Veksler, 2008; Singh, Saxena, & Jain, 2011; Vezhnevets, & Konouchine, 2005; Rother, Kolmogorov, & Blake, 2004; Li, Sun, Tang, & Shum, 2004). The previous initial segmentation methods achieve good results but suffer from complexity of object extraction from complex scene images. A linear time statistical region merging for fast segmentation is achieved in (Peng, Zhang, & Zhang, 2011).