Monte Carlo rendering graph denoising method based on generative adversarial network

Abstract

The invention relates to an image processing technology, discloses a Monte Carlo rendering image denoising method based on a generative adversarial network, and solves the problem of low image denoising efficiency caused by long network reasoning time in the prior art, and a denoising result can better recover low-frequency content and high-frequency details of a noise rendering image, so that the denoising efficiency is improved. Therefore, a more real de-noising result in vision can be obtained. According to the method, accurate and efficient denoising of the Monte Carlo rendered image is realized based on the constructed Monte Carlo rendered image denoising model, and the Monte Carlo rendered image denoising model is trained by the generative adversarial network; the architecture of the generative adversarial network comprises a de-noising network and an identification network, and the de-noising network is mainly composed of a noise feature encoder and an auxiliary feature encoder; the identification network is mainly composed of an identifier; the denoising network inputs the noise rendering graph and the auxiliary cache graph and outputs a denoising rendering graph; and the identification network is used for identifying true and false images of the input de-noised rendering image and the target rendering image corresponding to the noise rendering image.

Description

technical field [0001] The invention relates to image processing technology, in particular to a method for denoising a Monte Carlo rendered image based on a generation confrontation network. Background technique [0002] Monte Carlo path tracing rendering is a classic offline rendering technology. It uses a large number of samples to approximate the solution of the rendering equation by means of the Monte Carlo numerical estimation method. When the number of samples is large enough, the Monte Carlo numerical estimation method can guarantee an unbiased solution, making The final rendered image can converge to a theoretically real-world image. [0003] Specifically, the Monte Carlo rendering technology emits thousands of sampled rays from the camera to each pixel in the rendering plane, and each ray will bounce multiple times in the 3D scene until the ray bounces to the light source, but this leads to The convergence speed of this algorithm is very slow, although the speed ca...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is to propose a Monte Carlo rendering image denoising method based on generative confrontation network, which solves the problem of low image denoising efficiency due to long network reasoning time in the prior art, and the denoising result It can better restore the low-frequency content and high-frequency details of the noise rendering image to obtain a more visually realistic denoising result

Images