Dr.Ioannis IvrissimtzisAESHAH ALI ALMUTAIRI2022-06-042022-01-112022-06-04109593https://drepo.sdl.edu.sa/handle/20.500.14154/66452In computer graphics, triangle mesh has emerged as the ubiquitous shape rep- resentation for 3D modelling and visualisation applications. Triangle meshes, often undergo compression by specialised algorithms for the purposes of storage and trans- mission. During the compression processes, the coordinates of the vertices of the triangle meshes are quantised using fixed-point arithmetic. Potentially, that can alter the visual quality of the 3D model. Indeed, if the number of bits per vertex coordinate is too low, the mesh will be deemed by the user as visually too coarse as quantisation artifacts will become perceptible. Therefore, there is the need for the development of quality metrics that will enable us to predict the visual appearance of a triangle mesh at a given level of vertex coordinate quantisation. In this thesis, we present the results of four psychophysical experimental studies to assess the visual quality of quantised meshes. To achieve that, we used triangles meshes which varied in some important geometric characteristics, such as the number of triangles, the average shape of the triangles, and the level of detail in their global shape. The meshes in the experiments were quantised using di↵erent quantisation methods, rendered with the use of various textures and lighting environments. We employed various experimental designs, such as 2-AFC with or without staircases, and MLDS, and in all cases lightly trained participants were invited to compare the visual qualities of the models. The main findings of our experiments cab be summarised as follows. The dis- crimination threshold, that is, the level of quantisation below which the viewer can perceive quantisation artifacts, is lower when dithered quantisation is used instead of the most commonly used rounding. A large number of triangles in the mesh, and rendering methods with high specular components, increase the discrimina- tion thresholds, that is, they require more bits per vertex coordinate to make the quantisation artifacts imperceptible. In, perhaps, the strongest result in the thesis, we established a strong correlation between the discrimination threshold and the amount of information carried by the mesh, as measured by the file size of the mesh when compressed with a state-of-the-art method. A fourth experiment, based on a more complex design and the MLDS method, was not conclusive, but enabled a pre- liminary investigation on the challenges facing these types of complex experiments with lightly trained participants.118enThesis