High Dynamic Range for High Fidelity Image Synthesis of Real Scenes

计算机图形工业，特别是那些涉及电影、游戏、模拟、虚拟现实和军事应用的工业，继续要求更逼真的计算机生成图像，即更准确地匹配它们所要表现的真实场景的计算机图像。当考虑到自然世界的图像时，这尤其具有挑战性，因为自然世界呈现给我们的视觉系统的颜色和强度范围很广。在大多数真实场景中，例如，从房子里看向窗户，最暗的区域（例如房间内）和最亮的区域（窗外）之间的比率，即所谓的对比度，可能是数千比1。典型的计算机显示器的对比度仅为100:1左右，因此无法准确显示此类场景。为了在真实场景和计算机显示器上显示的内容之间实现感知匹配，已经开发了许多外观保持或色调映射操作符（TMOs）。然而，目前还不可能完全验证这些tmo与他们试图描绘的真实场景的保真度。最近发展的新型高动态范围（HDR）显示器，能够达到75,000:1的对比度，现在提供了计算和显示计算机生成的图像的机会，这些图像在感知上更接近现实世界。这项研究计划将使用这些新颖的HDR显示器来评估现有的TMOs，看看它们如何很好地保留了真实场景的外观，并将利用所获得的见解为现有的计算机显示器和HDR显示器开发新的，更精确的TMOs。还将制定一个框架，提供一种直接、客观的方法来比较真实图像和合成图像。两个应用，这是严重依赖于计算机图像的现实性，是虚拟考古和军事模拟。当在电脑上调查过去的环境时，无法产生与过去环境准确匹配的图像，实际上可能导致考古学家对过去的误解。类似地，军用车辆试图在特定地形中伪装的错误显示可能导致车辆在真实战场场景中被发现。我们将使用考古学和伪装的具体例子来检验我们的研究结果。

The computer graphics industry, and in particular those involved with films, games, simulation, virtual reality and military applications, continue to demand more realistic computer-generated images, that is computed images that more accurately match the real scene they are intended to represent. This is particularly challenging when considering images of the natural world, which presents our visual system with a wide range of colours and intensities. In most real scenes, for example, looking from inside a house towards a window, the ratio between the darkest areas (e.g. inside the room) and the brightest area (outside the window), the so-called contrast ratio, could be many thousands to one. A typical computer monitor only has a contrast ratio of about 100:1 and is thus incapable of accurately displaying such scenes.A number of appearance-preserving, or tone-mapping, operators (TMOs) have been developed in order to try to achieve a perceptual match between the real-world scene and what is displayed on the computer monitor. However, it has not yet been possible to validate the fidelity of these TMOs thoroughly against the real scenes they are trying to portray. The recent development of novel, high dynamic range (HDR) displays, capable of 75,000:1 contrast ratio now provide the opportunity to compute and display computer-generated images that are perceptually much closer to the real world.This research proposal will use these novel HDR displays to evaluate existing TMOs to see how well they do preserve the appearance of the real scenes, and will use the insights gained to develop new, more accurate TMOs for existing computer monitors and HDR displays. A framework will also be produced that will provide a straightforward, objective way of comparing real and synthetic images. Two applications, which are critically dependent on the realism of computed images, are virtual archaeology and military simulations. When investigating past environments on a computer, failure to produce images that accurately match what the past environment may have looked like, may in fact lead to the archaeologists misinterpreting the past. Similarly, the incorrect display of a military vehicle attempting to camouflage in a certain terrain may lead to detection of the vehicle in the real battlefield scenario. We will use specific examples from archaeology and camouflage to test the results of our research.