Progressive path tracing with bilateral-filtering-based denoising
- PDF / 4,652,444 Bytes
- 16 Pages / 439.642 x 666.49 pts Page_size
- 38 Downloads / 332 Views
Progressive path tracing with bilateral-filtering-based denoising Qiwei Xing1 · Chunyi Chen1 · Zhihua Li1 Received: 22 August 2019 / Revised: 12 August 2020 / Accepted: 18 August 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Path tracing can generate realistic images based on virtual 3D scene models, but the images are prone to be noisy. To solve this problem, we developed a novel denoising algorithm framework. Firstly, according to the relative mean square error of the noisy pixels, we introduced a progressive adaptive sampling strategy to optimize the distribution of samples. Next, to enhance the quality of the final reconstructed images, we designed an improved bilateral filtering algorithm with use of the gradient feature to obtain the noise-free images. Experimental results demonstrate that our framework outperforms the state-of-the-art path tracing denoising methods in terms of the visual quality, numerical error , and time cost. Keywords Path tracing · Image denoising · Progressive adaptive sampling · Gradient feature · Improved bilateral filtering
1 Introduction Path Tracing(PT) is among the most effective techniques for generating photorealistic images [10, 16].And in recent years, PT emerged as rendering algorithm of choice for film and visual effects [31]. For example, most of the high-end, film production pipelines now employ PT, and even real-time pipelines are moving towards physically-based rendering as modern games begun to use PT [10, 31]. Unless an excessive number of ray samples is distributed, PT renderings converge slowly and suffer from noise artifacts, i.e., variance at low sampling rates [19]. As a result, there is increasing demand to obtain the images generated by PT faster and with much fewer samples than before. Chunyi Chen
[email protected] Qiwei Xing [email protected] Zhihua Li [email protected] 1
School of Computer Science and Technology, Changchun University of Science and Technology, Changchun 130022, China
Multimedia Tools and Applications
Of all the variance reduction techniques proposed during recent years, PT denoising [12, 15, 29, 33],in particular, has helped to fuel the recent, rapid adoption of PT. PT denoisers for high-end images [1, 6, 25, 32, 36] have demonstrated impressive results at low sampling rates for their respective applications. Still, their results could substantially improve if they could make their input images better. Besides, they wish their denoising methods could be employed in real-time rendering systems [6, 25, 32]. Furthermore, their methods require a significant number of samples as the input. Then they denoise the input to generate the noise-free results. However, the more samples, the more time the algorithm takes [15, 25, 31, 32]. In addition, their reconstruction process are most based on the offline rendering systems, such as Physically Based Rendering Technique(PBRT) [3]. So, if we could improve the quality of input images and speed up the convergence, we can obtain the same quality at ever lower
Data Loading...
