Graph Spectral Image Processing

5Graph Spectral 3D Image Compression

Thomas MAUGEY¹, Mira RIZKALLAH², Navid MAHMOUDIAN BIDGOLI¹, Aline ROUMY1 and Christine GUILLEMOT¹

¹Inria Rennes – Bretagne Atlantique, Rennes, France

²École Centrale Nantes, France

In recent years, new types of image modalities have emerged. Most of them were developed in the context of immersive experience, such as virtual/augmented reality and 6 degree-of-freedom (6-DoF) visualization. These images tend to represent the three-dimensional (3D) world and are, therefore, referred to as 3D images. This implies that the dimension of the images captured with such devices is huge, and requires efficient compression algorithms. Another specificity of these new images is that their shapes differ from the traditional two-dimensional (2D) array of pixels, which, in turn, implies that most of the existing compression algorithms are not directly adapted. Generally, these algorithms are used after some mappings onto 2D planes, which may lead to sub-optimality. As explained in the previous chapters, graph signal processing techniques are able to handle irregular topology, and thus become a natural tool for compressing these new images efficiently.

This chapter is organized into three main parts: section 5.1 presents an introduction to the different 3D image modalities considered in the chapter; section 5.2 presents the overall graph-based coding scheme of a 3D image and section 5.3 details the different graph construction techniques.

5.1. Introduction ...