Point Cloud Quality Assessment

Nowadays, richer 3D visual representation formats are emerging, notably light fields and point clouds. These formats enable new applications in many usage domains, notably virtual and augmented reality, geographical information systems, immersive communications, and cultural heritage. Recently, following major improvements in 3D visual data acquisition, there is an increasing interest in point-based visual representation, which models real-world objects as a cloud of sampled points on their surfaces. Point cloud is a 3D representation model where the real visual world is represented by a set of 3D coordinates (the geometry) over the objects with some additional attributes such as color and normals. With the advances in 3D acquisition systems, it is now possible to capture a realistic point cloud to represent a visual scene with a very high resolution. These point clouds may have up to billions of points and, thus, storing and transmitting them in a raw format would require an unbearable amount of memory and bandwidth. Therefore, the storage and transmission of large point clouds critically ask for the development of efficient point cloud coding solutions. In this context, to boost a wide adoption of this 3D visual representation model, it is also necessary to reliably measure the quality of experience offered to the end-users by measuring the point cloud quality. While objective quality assessment metrics aim to mathematically measure the quality of point clouds, notably decoded point clouds, ideally replicating the scores that would be given by human beings, the subjective quality assessment allows not only to perform more reliable assessment but also allows to assess the correlation of the available objective quality metrics with the users? opinion scores. The design and identification of the most reliable objective quality metrics, notably for point clouds, requires subjective evaluation data obtained in meaningful and already proven methodology. In this context, the main objective of this Thesis is twofold: first, to perform appropriately designed subjective quality assessment experiments for decoded point clouds under different degradations and impacting factors like coding and rendering which allows to assess and benchmark the reliability of point cloud objective quality metrics; and second, to propose novel objective quality metrics with a higher correlation with the obtained subjective assessment scores. To achieve these objectives, three subjective quality assessment experiments have been performed considering different contents, degradations, and impact factors. Moreover, four objective quality metrics have been proposed, all outperforming the state-of-the-art objective quality metrics at the time they were developed. Due to the importance of geometry on point cloud perceived quality, and new challenges associated with geometry quality evaluation, most of the efforts in this Thesis were around geometry quality evaluation, notably for static point clouds. However, in the last chapter, a quality metric jointly considering geometry and color is proposed, which outperforms all available quality metrics in the literature for point clouds.