High quality texture synthesis

Texture synthesis is a core process in Computer Graphics and design. It is used extensively in a wide range of applications, including computer games, virtual environments, manufacturing, and rendering. This thesis investigates a novel approach to texture synthesis in order to significantly improve speed, memory requirements, and quality. An analysis of texture properties is created, to enable the gathering a representative dataset, and a qualitative evaluation of texture synthesis algorithms. A new algorithm to make non-repeating texture synthesis on-the-fly possible is developed, tested, and evaluated. This parallel patch-based method allows repeatable sampling without cache, without creating visually noticeable repetitions, as confirmed by a perceptive objective study on quality. In order to quantify the quality of existing algorithms and to facilitate further development in the field, desired texture properties are classified and analysed, and a minimal set of textures is created according to these properties to allow subjective evaluation of texture synthesis algorithms. This dataset is then used in a user study which evaluates the quality of texture synthesis algorithms. For the first time in the field of texture synthesis, statistically significant findings quantify the quality of selected repeatable algorithms, and make it possible to evaluate new improved methods. Finally, in an effort to make these findings applicable in the British tile manufacturing industry, the developed texture synthesis technology is made available to Johnson Tiles