Generación de imágenes fotorrealistas de alta resolución con la técnica de upsampling

Abstract

[en] The rendering of a 3D virtual scene to a photorealistic image is a procedure with a high computational complexity, it depends on various factors, one of the most important being illumination, simulating light rays, the number of bounces they can have on the surfaces of 3D models, how this will affect the scene, among other things, requires many calculations, to arrive at a correct approximation, this is directly reflected in high execution times for the generation of quality images. One of the current solutions to generate quality photorealistic images is the path tracing rendering algorithm, which, being based on the Montecarlo method, will denote that the quality of its results is directly proportional to the number of samples (the simulated light rays per pixel), requiring an extremely high number of samples to achieve a quality photorealistic image. Given this problem, during the development of this Final Degree Project, various upsampling techniques have been developed, with the basic procedure that working with an image generated at lower resolution by path tracing, using different metrics to obtain the best possible interpolation, to have as a final result, a high-resolution photorealistic image with equal or superior quality in less time than the path tracing could take. The algorithms developed here expose different theories about why that metric and/or interpolation supposes an improvement, seeking that each new proposed algorithm complements or covers aspects not contemplated by the previous one. The results obtained with the virtual 3D scenes created, compared to path tracing, allow us to understand the operation of the algorithm while demonstrating that with the appropriate parameters depending on the algorithm used, the quality photorealistic image is achieved in less execution time.