A method for restoring the texture of equiluminous objects in a scene

S. I. Kupriyanov, D. D. Zhdanov, I. V. Valiev


Abstract: The article presents a method for restoring the texture of equiluminous (diffuse) surfaces of scene objects. This method is supposed to be used as an element of data preprocessing in the task of restoring the parameters of the optical properties model by means of differentiable rendering. Restoring the texture of diffuse surfaces of a scene involves creating a UV scan of the geometry onto a plane, forming a texture brightness map using the projection of cameras onto the texture plane, and selecting the components of a diffuse optical properties model based on the brightness and illumination of the scene. The created brightness map and original images, combined with the parameters of the geometry and optical properties of the scene, make it possible to calculate the primary lighting of the scene and select the secondary one. Based on this data, you can calculate the color of the surface. The optimization process starts calculating secondary illumination using direct ray tracing while maintaining illumination and brightness on textured objects. The secondary illumination and global illumination brightness obtained during iterative calculations are compared with the initial brightness of the image, and the presented algorithm corrects the color of the restored surface and re-evaluates the new distribution of secondary illumination. The process is repeated until the brightness deviation of the received image from the brightness of the original image is within the specified limits. The proposed approach requires a pre-defined geometry of the scene, the parameters of the light sources, as well as the original images for the specified camera parameters. The restored color of the texture of the scene objects allows it to be used as the initial data of the scene parameters in the differentiable rendering system. In addition, the found chromaticity of scene objects makes it possible to simplify the process of differentiable rendering by eliminating strong brightness variations and reducing the number of points in the scene image transmitted to differentiable rendering.

Keywords: differentiable rendering, optical properties, image preprocessing, direct illumination, secondary illumination geometry analysis, brightness, color, geometry scanning, texturing, optimization.