Graph Spectral Image Processing

8Graph Spectral Image Segmentation

Michael N

G The University of Hong Kong, China

8.1. Introduction

Image segmentation is an important and fundamental step in computer vision, image analysis and recognition (Gonzales and Woods 2018). It refers to partitioning an image into different regions, where each region has its own meaning or characteristic in the image (e.g. the same color, intensity or texture). In the literature, there are a large number of image segmentation methods, including threshold-based, edge-based, region-based and energy-based approaches; see the references in Peng et al. (2013).

These approaches have been applied to many image processing applications successfully, for example, in medical imaging, tracking and recognition. For image segmentation methods, the energy-based approach is to develop and study an energy function, which gives an optimum when the image is segmented into several regions, according to the objective function criteria. This approach includes several techniques, such as active contour (for example, Kass et al. (1988)) and graph cut (for example, Shi and Malik (2000), Boykov et al. (2001)). The main advantage of using graph cut is that the associated energy function can be globally optimized, whereas this may not be guaranteed in the other segmentation methods. In the graph cut segmentation, the energy function is constructed based on graphs where image pixels are mapped to graph vertices, and it can be optimized via graph-based algorithms ...