TONE MAPPING OPTIMIZATION FOR REAL TIME APPLICATION
Date
2019-08-26T18:26:56Z
Authors
Zhao, Siwei
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Manually tuning tone mapping parameters is difficult, especially in real time applications. In this thesis, we propose modifications to an automated parameter tuning algorithm which can replace the manual tuning work using a content-adaptive tone mapping operator specially designed for rendered HDR content characteristics. We show that, our tone mapping results with optimized parameters preserve more visible contrast than other optimization algorithms. We also present a further algorithm for real-time interpolation of multiple look-up tables in video games accelerated by computing shaders on the GPU. Our method utilizes compute shaders to parallel counting the pixels belonging to each LUT and change the color grading effects using a post processing volume. We show that our results have a closer visual appearance to the reference images when the player changing the view point and can be executed with a fluent frame rate without overly taxing the CPU.
Description
Many tone mapping algorithms for natural and synthetic images have been proposed in recent years. However, manually tuning their parameters is difficult, especially if the scene is dynamically changing over time or the view point is altered, as is the case in real time applications. In this thesis, we propose modifications to an automated parameter tuning algorithm which can replace the manual tuning work. This algorithm can optimize the parameters of tone mapping operators by minimizing the perceptual distortion using an evolution strategy. The perceptual distortion is measured by a feature-based objective image quality metric. This metric utilizes a contrast-enhanced virtual photograph technique by using a content-adaptive tone mapping operator specially designed for rendered HDR content characteristics. We show that, our tone mapping results with optimized parameters preserve more visible contrast than other optimization algorithms. A further related issue in real time applications is the interpolation of tone mapping parameters. Standard practice uses a color grading pipeline to encode tone mapping results and color grading modifications into look-up tables (LUTs). Using look-up tables, artists can reproduce and apply certain luminance and color grading processes efficiently and easily. However, the interpolation of tone mapping and color grading parameters between locations where look up tables have been designed is problematic. We present a further algorithm for real-time interpolation of multiple look-up tables in video games accelerated by compute shaders on the GPU. Our method utilizes compute shaders to parallel counting the pixels belonging to each LUT and change the color grading effects using a post processing volume. We show that our results have a closer visual appearance to the reference images when the player changing the view point and can be executed with a fluent frame rate without overly taxing the CPU.
Keywords
Tone Mapping, High Dynamic Range, Games