HDR成像技术：从捕捉到显示的革新

book

需积分: 5 103 浏览量更新于2024-06-30 收藏 19.28MB PDF 举报

身份认证购VIP最低享 7 折!

领优惠券(最高得80元）

"高动态范围成像 (HDR) 技术在SIGGRAPH2004课程中的探讨，由Paul Debevec、Erik Reinhard、Greg Ward和Sumanta Pattanaik等专家主讲，该课程涵盖了从捕捉到显示的HDR成像最新进展。" 高动态范围成像（HDR）是一种摄影和图像处理技术，它能够捕捉和存储比传统图像格式更广泛的亮度和色彩范围。这一技术的出现解决了普通显示设备对比度和颜色显示范围有限的问题，从而使得图像能更真实地再现原始场景的色彩深度和动态范围。在SIGGRAPH 2004的这门课程中，专家们详细讲解了HDR成像的各个环节，包括： 1. **图像捕捉**：使用特殊的相机设置和传感器，可以捕捉超出标准8位色彩通道的亮度信息。这通常涉及多曝光拍摄，通过不同曝光时间的几张照片组合来获得更广阔的动态范围。 2. **文件格式**：HDR图像需要特殊格式来存储额外的亮度信息，如 Radiance RGBE（辐射亮度）格式或OpenEXR格式。这些格式允许保留超出8位或16位的精度，以保存高亮和阴影细节。 3. **图像处理**：处理HDR图像涉及到将高动态范围数据转换为适合低动态范围（LDR）显示设备的图像。这包括色调映射（tone mapping），一种压缩HDR信息以适应有限显示范围的技术。不同的色调映射算法会产生不同的视觉效果，如对比度增强、细节保留等。 4. **显示考虑**：由于大多数显示器目前仍无法直接显示HDR，课程还讨论了如何将HDR图像适配到这些设备，包括色彩空间转换、亮度限制和对比度调整等步骤。 5. **实践应用**：课程提供实践经验，教导学员如何在实际操作中应用这些技术，从拍摄到后期处理，再到最终在LDR显示器上呈现。 6. **权衡分析**：在从捕捉到显示的每个阶段，都有其技术挑战和权衡。例如，高曝光可能导致噪声增加，而色调映射则可能牺牲某些细节或改变图像的自然外观。通过这门课程，参与者可以深入了解HDR成像的理论与实践，提升对图像质量的理解，并掌握在现代数字媒体中应用HDR的关键技能。这种技术不仅在电影制作、游戏开发、虚拟现实等领域有广泛应用，也逐渐渗透到日常的摄影和图像处理中，为创造更为逼真和震撼的视觉体验提供了可能。

资源详情

资源推荐

“imaginary” primaries, as in the CIE XYZ color system, can encode the visible gamut with positive

values, but often at the expense of coding efficiency since many unreal colors will also be represented –

similar to the useless dynamic range issue. Finally, the distribution of error is not perceptually uniform

with this encoding. In particular, step sizes may become visible in saturated blue and magenta regions

where the green mantissa drops below 20. These shortcomings led us to develop a much better format,

described next.

SGI LogLuv (TIFF)

Working at SGI in 1997, we set about correcting the mistakes made with RGBE, in hopes of providing

an industry standard for HDR image encoding. This ultimately resulted in a LogLuv codec in Sam

Leffler’s TIFF library. This encoding is based on visual perception, and designed such that the

quantization steps match human contrast and color detection thresholds (a.k.a. “just noticeable

differences”). The key advantage is that quanta in the encoding are held below the level that might result

in visible differences on a “perfect” display system. Its design is similar in spirit to a YCC encoding, but

with the limitations on color gamut and dynamic range removed. By separating the luminance and

chrominance channels, and applying a log encoding to luminance, we arrive at a very efficient

quantization of what humans are able to see. This encoding and its variants were described in the 1998

Color Imaging Conference paper, “Overcoming Gamut and Dynamic Range Limitations in Digital

Images.”

There are actually three variants of this logarithmic encoding. The first squeezes a 10-bit log luminance

value together with a 14-bit CIE (u´,v´) lookup to fit everything into a standard-length 24-bit pixel. This

was mainly done to prove a point, which is that following a perceptual model allows you to make much

better use of the same number of bits. In this case, we were able to extend to the full visible gamut and

4.8 orders of magnitude of luminance in

just

imperceptible steps. The second variation uses 16 bits for a

ure luminance encoding, allowing negative values and covering a dynamic range of 38 orders of

magnitude in 0.3% steps, which are comfortably below the perceptible level. The third variation uses the

same 16 bits for signed luminance, then adds 8 bits each for CIE u´ and v´ coordinates to encompass all

the visible colors in 32 bits/pixel.

Although the LogLuv format has been adopted by a number of computer graphics researchers, and its

incorporation in Leffler’s TIFF library means that a number of programs can read it even if they don’t

know they can, it has not found the widespread use we had hoped. Part of this is people’s reluctance to

deviate from their familiar RGB color space. Even something as simple as a 3

3 matrix to convert to

and from the library’s CIE XYZ interface is confounding to many programmers. Another issue is simply

awareness – unless one has key industry partners, it is difficult to get word out on a new format, no

matter what its benefits might be. The follow-up article, written for the ACM Journal of Graphics Tools

(vol. 3, no. 1), “LogLuv encoding for full-gamut, high-dynamic range images,” was an effort to stem

these difficulties. We are still hopeful that this format will find wider use, as we have found it to work

extremely well for HDR image encoding, ourselves. There is no more appropriate format for the

archival storage of color images, at least until evolution provides an upgrade to human vision.

ILM OpenEXR (EXR)

More recently, Industrial Light and Magic published C++ source code for reading and writing their

OpenEXR image format, which has been used internally by the company for special effects rendering

and compositing for a number of years. This format is a general-purpose wrapper for the 16-

it Half data

type, which has also been adopted by both NVidia and ATI in their floating-point frame buffers.

Authors of the OpenEXR library include Florian Kainz, Rod Bogart (of Utah RLE fame), Drew Hess,

Josh Pines, and Christian Rouet. The Half data type is a logical contraction of the IEEE-754 floating

oint re

resentation to 16 bits. It has also been called the “S5E10” format for “Si

lus 5 ex

onent

lus

e 9 of 25HDR Ima

e Encodin

10 mantissa,” and this format has been floating around the computer graphics hardware developer

network for some time. (OpenEXR also supports a standard IEEE 32-bit/component format, which we

will not discuss here.)

Because it can represent negative primary values along with positive ones, the OpenEXR format covers

the entire visible gamut and a range of about 10.7 orders of magnitude with a relative precision of 0.1%.

Since humans can simultaneously see no more than 4 orders of magnitude, this makes OpenEXR a good

candidate for archival image storage. Although the basic encoding is 48 bits/pixel as opposed to 32 for

the LogLuv and RGBE formats, the additional precision is valuable when applying multiple blends and

other operations where errors might accumulate. It would be nice if the encoding covered a larger

dynamic range, but 10.7 orders is adequate for most purposes, so long as the exposure is not too

extreme. The OpenEXR specification offers the additional benefit of extra channels, which may be used

for alpha, depth, or spectral sampling. This sort of flexibility is critical for high-end compositing

ipelines, and would have to be supported by some non-standard use of layers in a TIFF image.

We predict that the OpenEXR format will find widespread use in the computer graphics research

community and special effects industry, as it has clear advantages for high quality image processing, and

is supported directly by today’s high-end graphics cards. ILM is to be commended for offering their

excellent library for license-free use, and given the complexity of the most common variant, PIZ lossless

wavelet compression, reimplementing EXR in a private library seems a formidable task.

Microsoft/HP scRGB Encoding

A new set of encodings for HDR image representation have been proposed by Microsoft and Hewlett-

Packard, and accepted as an IEC standard (61966-2-2). The general format is called scRGB, formerly

known as sRGB64. This standard grew out of the sRGB specification developed by Michael Stokes,

Matthew Anderson, Srinivasan Chandrasekar, and Ricardo Motta, at HP and Microsoft. In essence, it is

a logical extension of 24-bit sRGB to 16 linear bits per primary, or 12 bits per primary using a gamma

encoding.

The scRGB standard is broken into two parts, one using 48 bits/pixel in an RGB encoding and the other

employing 36 bits/pixel either as RGB or YCC. In the 48 bits/pixel substandard, scRGB specifies a

linear ramp for each primary. Presumably, a linear ramp is employed to simplify graphics hardware and

image-processing operations. However, a linear encoding spends most of its precision at the high end,

where the eye can detect little difference in adjacent code values. Meanwhile, the low end is

impoverished in such a way that the effective dynamic range of this format is only about 3.5 orders of

magnitude – not really adequate from human perception standpoint, and too limited for light probes

and

HDR environment mapping. The standard does allow for negative primaries, which is an improvement

over earlier RGB standards, but it makes wasteful use of this capability as much of the range is given to

representing imaginary colors outside of the visible gamut, especially at the bottom end of the scale,

where human color perception actually degrades. At the top end of its dynamic range, where people see

colors more clearly, the gamut collapses in on itself again as primaries get clipped to the maximum

representable value. This can be visualized on the colorspace animation page

The second part of the standard offers a more sensible and compact 36 bits/pixel specification, which

employs a standard gamma ramp with a linear subsection near zero. Although it uses 25% fewer bits, it

has nearly as great a dynamic range as the 48-bit version, 3.2 orders as opposed to 3.5, and also allows

negative primary values. It still suffers from a collapsing of its gamut at the top end and the unusable

colors at the bottom, but it’s a better encoding overall. Even better than this scRGB-nl substandard is the

scYCC-nl encoding, which benefits from a separate luminance channel and a color space that does not

collapse quite as quickly at the top end. However, there is some strange behavior at peak luminance

where the bri

htest re

resentable value is not white, but follows a rin

artiall

saturated colors

e 10 of 25HDR Ima

e Encodin

around white. This is because Cb and Cr contribute to the final luminance, so colored values (where Cb

and Cr are non-zero) can be brighter than neutral ones. The net result is that colored light sources can

become more saturated rather than less saturated during standard gamut clamping – fading to hot pink,

for example. This situation is illustrated in Figure 4, below. This may also be seen on the colorspace

animation page. In most situations, this would be an unexpected result, so care should be exercised when

erforming gamut mapping into scYCC-nl.

Figure 4.

Gamut clamping anomalies using the scYCC-nl encoding. The first image shows a linear tone-mapping with

clamping on the original HDR image of the sun. The second image shows the effect of simple clamping into an scYCC-nl

gamut. Strange colorations result from the fact that the maximum representable luminances in scYCC-nl are not white.

It is unclear where the scRGB standard is headed at this point. Microsoft is promoting an undisclosed

variant of this standard for use in digital cameras, graphics cards, and their Longhorn presentation

engine, code named “Avalon” (October 2003 MS Developer’s Conference). In the case of digital

cameras, the only competing high resolution standard is no standard at all – the so-called “RAW” image

formats that are not only manufacturer-specific, but camera model-specific as well. This is a disaster

from a user and software maintenance standpoint, and almost any data standard would be an

improvement over the current situation. Modern graphics cards, on the other hand, have already adopted

the Half floating point data type used in OpenEXR and the two leading manufacturers, NVidia and ATI,

have designed their chips around it. There would seem to be little benefit in switching to a more limited

integer standard such as scRGB at this point. As for image processing software, the only benefit over

existing 48-bit integer RGB formats is the extended gamut, which comes at the expense of dynamic

range. A 48-bit RGB pixel using a standard 2.2 gamma as found in conventional TIFF images holds at

least 5.4 orders of magnitude, though applications like PhotoshopCS are not presently designed to use

this range to best advantage. Image processing would probably be done in memory using 32-bit IEEE

floats, and the scRGB standard would only be useful for file storage. However, scRGB and scRGB-

nl/scYCC-nl hold less dynamic range than any of the other formats we have discussed for this purpose.

We hope that the secret variant that Microsoft is promoting is a substantial improvement over the written

IEC standard.

HDR Encoding Comparison

The following table summarizes the information given in the previous section. The first row entry in the

chart shows 24-

it sRGB, which of course is not a hi

h d

namic-ran

e standard, but offered here as a

e 11 of 25HDR Ima

e Encodin

剩余275页未读，继续阅读

慢吞吞的倒霉熊

粉丝: 1
资源: 16

HDR成像技术：从捕捉到显示的革新

High Dynamic Range Imaging.pdf

High Dynamic Range Imaging - Acquisition, Display, and Image-Based Lighting

High-Dynamic-Range Displays

high dynamic range imaging and tone mapping marc alexa

dynamic range imaging pdf

HIS颜色模型中的HIS指的是

请你给我一些关于深度学习进行影像位深降位的资料和链接和作者，要求需要很准确

realisticVision

burst photography for high dynamic range and low-light imaging on mobile cam

unity HDRI

Surreal Capture

a= ['magnetic', 'resonance' ,'imaging']使用循环得到b =['imaging','resonance imaging']

the early second paper about Optical computational imaging

a= ['magnetic', 'resonance' ,'imaging']依次取出 imaging,resonance imaging

[('magnetic', 'JJ'), ('resonance', 'NN'), ('imaging', 'NN')]对其进行操作，取出'resonance imaging'和'imaging'这两个名词

commons-imaging依赖

multi-beam imaging matlab

from . import _imaging as core

Java Advanced Imaging怎么安装

aspose.imaging for net去除水印

最新资源