Assessing the quality of 3D point clouds using descriptors for color and geometry texture

Abstract

Since the mid-20th century, the use of digital formats for visual content has revolutionized communication in society. The Internet and digital broadcasting systems, which became widely available in the 1990s, led to an incredible expansion of multimedia consumption among the public. As a result, telecommunication networks and providers were pushed to their limits to meet the growing demand for multimedia content. Traditional electronic imaging systems, such as TV broadcasting systems, were designed based on subjective quality analysis for defining parameters like the number of lines in a video. However, with the emergence of new digital visual content services that require faster and more affordable methods of evaluating human perceived quality, there has been a need for automatic quality assessment. In the past few decades, numerous visual quality models based on algorithms running on digital computers have been proposed. While these existing models are highly advanced for 2D digital imagery, a new generation of immersive media has emerged with different data structures that are not applicable to traditional 2D methods and require novel quality assessment metrics. These emerging immersive media formats provide a 3D visual representation of real objects and scenes. In this new visual format, objects can be captured, compressed, transmitted, an visualized in real-time as 3D content, allowing consumers to select free-viewpoints to view such media. One of the most popular formats for immersive media is Point Cloud (PC), which is composed of points with 3D geometry coordinates and color information, and sometimes other attributes such as reflectance and transparency. This research focuses on the quality assessment of 3D Point Clouds using novel color and geometric texture statistics. Since distortions in both color and geometry attributes of 3D visual content can affect the perceived visual quality, this work proposes using both color-based and geometry-based texture descriptors for Point Clouds to obtain visual degradation information through their statistics. Four novel Point Cloud texture descriptors are introduced in this work, three of which are color-based and one which is geometry-based. Additionally, a new voxelization method is proposed, which converts points to voxels (volume elements), and improves the performance of the color-based texture descriptors. The proposed Point Cloud quality assessment method demonstrates high performance, comparable to the state-of-the-art methods, while also being flexible and extensible to adapt to different types of distortions.