{"slug":"topaz-denoise-high-iso","title":"7万高ISO也能出大片？这个降噪技术太厉害了！","author":"Thomas看看世界","date":"2020-08-02","wechat":"https://mp.weixin.qq.com/s?__biz=MzI4NDA5MzE5NQ==&mid=2247491089&idx=1&sn=6b4231becfc0a14d661d1db0d63fbd54&chksm=eb81e075dcf66963ab98d105bf8e4afa1267e7a25ceaf5784dba123dedfdbb1f29c3e1e2735f&scene=21#wechat_redirect","blocks":[{"t":"h","x":"Thomas看看世界","xe":"Thomas Chu Photo"},{"t":"p","x":"本文是一篇关于高ISO降噪的专业教程，作者是著名野生动物摄影师杰夫。","b":false,"xe":"This is a professional tutorial on high-ISO noise reduction, written by renowned wildlife photographer Jeff."},{"t":"p","x":"掌握这套降噪流程之后，即使上万ISO的照片也能有效降噪。本文字数近11000，建议大家先转发+收藏！","b":false,"xe":"Once you master this workflow, even photos shot above ISO 10,000 can be effectively denoised. This article runs nearly 11,000 words — share and bookmark it first!"},{"t":"img","f":"img-02","w":1080,"h":660},{"t":"p","x":"野生动物摄影、体育摄影、纪实摄影，包括很多婚纱摄影师在室内低光环境中拍摄婚礼现场时，不得不拍摄ISO很高的图像。","b":false,"xe":"Wildlife photography, sports photography, documentary photography — and many wedding photographers shooting indoor low-light ceremonies — all have no choice but to shoot at very high ISO."},{"t":"p","x":"以野生动物摄影来说，大多数野生动物最活跃的时候，是日出日落前后的半个小时，尤其是拍摄金钱豹这种全夜行的猫科动物，拍摄时间一定是在低光，它又喜欢潜伏在树丛里，所以拍摄环境总是非常暗。","b":false,"xe":"Take wildlife photography: most wild animals are most active in the half hour around sunrise and sunset. Especially for a fully nocturnal cat like the leopard, shooting time is always in low light, and it likes to lurk in the bushes — so the shooting environment is always very dark."},{"t":"p","x":"而为了拍摄高速动态场景，比如捕猎打斗，我们需要起码1/1600s或更快的快门速度来清晰地冻结动物的动作，F5.6、F8或更小的光圈来获得足够的景深，以使得即使不在一个焦平面的两个主体都清晰。这样我们要获得足够感光度的唯一办法是提高ISO。","b":false,"xe":"To capture high-speed action like hunting and fighting, we need at least 1/1600s or faster to freeze the animals' movement, and F5.6, F8, or smaller apertures for enough depth of field so that two subjects on different focal planes both stay sharp. The only way to get enough sensitivity is to raise the ISO."},{"t":"img","f":"img-03","w":1080,"h":608},{"t":"p","x":"我拍摄野生动物的曝光模式设置是M档，快门速度1/1600s以上，光圈F5.6以上，ISO自动不设上限。这样，ISO的数值经常是10000以上，会充斥着噪点，所以我研究出了一个高ISO图像的拍摄以及后期处理工作流程——“杰夫流程”，来处理这些有很多噪点的图像，以达到可以使用的程度。","b":false,"xe":"My exposure settings for wildlife: Manual mode, shutter 1/1600s or faster, aperture F5.6 or smaller, Auto ISO with no upper limit. ISO often exceeds 10,000 and the images are riddled with noise, so I developed a shooting and post-processing workflow for high-ISO images — the \"Jeff Workflow\" — to bring these noisy files up to a usable standard."},{"t":"p","x":"比如下面我在2018年在马赛马拉拍摄的一个猎豹捕猎的场景，发生在日落之间的树林里，光线已经非常暗了，这张照片的ISO是14400,用“杰夫流程”处理后达到了专业摄影的要求，发表在2019年2月6日英国《泰晤士报》上：","b":false,"xe":"For example, this cheetah hunt I shot in the Maasai Mara in 2018 took place in the woods at sundown, when the light was already very dim. This photo was shot at ISO 14,400; processed with the \"Jeff Workflow,\" it met professional standards and was published in The Times (UK) on February 6, 2019:"},{"t":"img","f":"img-04","w":750,"h":934},{"t":"h","x":"1. 为什么要更新“杰夫流程”","xe":"1. Why Update the \"Jeff Workflow\""},{"t":"p","x":"从2014年我发布了“杰夫流程—高ISO，高速摄影噪点控制工作流程”。到现在已经6年了；这六年里，“杰夫流程”在各个摄影网站，我的书和教学视频在网络的传播已经超过50万次，我也收到了很多得到这个流程帮助的摄友的反馈，能够帮助到广大摄影师们我很高兴，这也是我当初把它放在网络上传播最主要的原因。","b":false,"xe":"Six years have passed since I published \"The Jeff Workflow — Noise Control for High-ISO, High-Speed Photography\" in 2014. In those six years, the workflow has spread over 500,000 times across photography websites, my books, and my teaching videos, and I've received much feedback from photographers it has helped. I'm glad it has helped so many — that was the main reason I put it online in the first place."},{"t":"p","x":"后期流程是帮助完善摄影师创作的工具，这些后期的工具一定会在这个技术高速发展的时代很快过时，“杰夫流程”的主要技术概念——用“向右曝光”的方法来尽可能多地记录获得图像细节，然后用软件来除去噪点并尽可能保持住画质这个做法没有改变，这是基于数码感应器记录数据的物理特性决定的。","b":false,"xe":"A post-processing workflow is a tool to help perfect a photographer's work, and such tools inevitably become outdated in an era of rapid technological development. The Jeff Workflow's core concept — using \"expose to the right\" to record as much image detail as possible, then removing noise in software while preserving quality — has not changed, because it is dictated by the physical properties of how digital sensors record data."},{"t":"img","f":"img-05","w":1080,"h":719},{"t":"p","x":"这六年来，除噪软件在不断进步，但是由于噪点像素形成会不可避免地占用、影响图像实际纹理像素的质量，从而影响到图像质量，所有的除噪软件的完善都是在更加精细地区分这些像素，在技术上并没有一个根本的突破。","b":false,"xe":"Over these six years, noise-reduction software has kept improving. But because noise pixels inevitably occupy and degrade the actual texture pixels, all such software improvements have been about distinguishing those pixels more precisely — there has been no fundamental technical breakthrough."},{"t":"p","x":"由于使用“杰夫流程”的关系,我一直在网上注意新的降噪软件的出现并试用，这几年我分别使用过的软件有：","b":false,"xe":"Because I use the Jeff Workflow, I've kept watching for and trying new denoise software. Over the years I've used:"},{"t":"p","x":"Noiseware","b":false,"xe":"Noiseware"},{"t":"p","x":"Skylum Luminar","b":false,"xe":"Skylum Luminar"},{"t":"p","x":"Nik Dfine 2","b":false,"xe":"Nik Dfine 2"},{"t":"p","x":"Noise Ninja","b":false,"xe":"Noise Ninja"},{"t":"p","x":"Neat image pro","b":false,"xe":"Neat image pro"},{"t":"p","x":"DxO Optics Pro 11 elite","b":false,"xe":"DxO Optics Pro 11 elite"},{"t":"p","x":"Topaz DeNoise AI","b":false,"xe":"Topaz DeNoise AI"},{"t":"p","x":"Topaz Denoise 是一款旧的软件，去年它出了人工智能AI版，马上引起了我的注意，刚刚试用时它的效果并不好，经常犯一些特别“二”的错误。","b":false,"xe":"Topaz DeNoise is an old piece of software; last year it released an AI version that immediately caught my attention. When I first tried it, the results weren't good — it often made particularly silly mistakes."},{"t":"p","x":"但是现在AI软件的优势是可以通过不断升级的方式快速调整、完善算法，到今年（2020年）3月份的这次升级，它的效果终于让我惊艳并决定使用人工智能来进行高ISO图像的处理，因为AI技术的插值功能和旧的降噪软件有本质的不同，因此新的“杰夫流程”也必须做一个相应的改变。","b":false,"xe":"But the advantage of AI software is that it can rapidly adjust and improve its algorithms through continuous upgrades. With the March 2020 upgrade, its results finally amazed me, and I decided to use AI for high-ISO image processing. Because AI's interpolation capability is fundamentally different from old denoise software, the new \"Jeff Workflow\" had to change accordingly."},{"t":"h","x":"2. 人工智能降噪软件相对于其他降噪软件的区别","xe":"2. How AI Denoise Software Differs from Other Denoisers"},{"t":"p","x":"在技术上，人工智能降噪软件相对于其他传统降噪软件进行全画面均匀降噪方法的本质区别是DRIR即“侦测（Detection）”、“识别(Recognition）”、“插值（Interpolation）”和“再渲染（Re-rendering）”。","b":false,"xe":"Technically, the essential difference between AI denoise software and traditional whole-image uniform denoising is DRIR: Detection, Recognition, Interpolation, and Re-rendering."},{"t":"p","x":"“侦测”是指对于不清晰的像素对比度的分析来对焦内焦外、和边缘的定位。","b":false,"xe":"\"Detection\" is the analysis of unclear pixel contrast to locate what's in focus, what's out of focus, and where the edges are."},{"t":"p","x":"“识别”是对图像的内容进行理解，并比较其数据库里的内容，进行判断图像的纹理逻辑推定。","b":false,"xe":"\"Recognition\" is understanding the image content and comparing it against a database to make logical inferences about the image's textures."},{"t":"p","x":"“插值”是在“侦测”和“识别”的基础上，是对画面里不清楚（特别是焦距外和噪点多的地方）的内容进行解读，对图像画质差的部分重新定义安排像素，并根据内容进行不同程度的降噪和画面细节修补的计算。","b":false,"xe":"\"Interpolation,\" building on detection and recognition, interprets the unclear parts of the frame (especially out-of-focus and noisy areas), redefines and rearranges pixels in low-quality regions, and computes varying degrees of noise reduction and detail repair based on content."},{"t":"p","x":"“再渲染”是根据“插值”计算完的结果重新安排像素，渲染形成新的图像。","b":false,"xe":"\"Re-rendering\" rearranges the pixels according to the interpolation results, rendering a new image."},{"t":"p","x":"下面我们会用实例说明新的“杰夫流程”怎样利用这些功能来对充斥着噪点的高ISO图像进行降噪和画质控制。","b":false,"xe":"Below we'll use real examples to show how the new \"Jeff Workflow\" leverages these capabilities to denoise high-ISO images and control image quality."},{"t":"p","x":"“杰夫流程2.0” 第一步：向右曝光（ETTR）","b":false,"xe":"\"Jeff Workflow 2.0\" Step 1: Expose to the Right (ETTR)"},{"t":"p","x":"敲黑板说重点——这一步的关键是：直方图右边不要被切断（除非是有目的的）以防止高光溢出。","b":false,"xe":"Key point — the crucial thing in this step: the right side of the histogram must not be cut off (unless intentional), to prevent highlight clipping."},{"t":"p","x":"自从2003年摄影师麦克尔.瑞克曼（Michael Reichmann）的文章“向右曝光（EXPOSE RIGHT）”开始，12年来网上有无数文章讨论“向右曝光”这个理论的科学依据，有兴趣的摄友可以自己去搜一些回来看。我先简单地说一下什么是向右曝光：","b":false,"xe":"Since photographer Michael Reichmann's 2003 article \"Expose Right,\" countless articles over 12 years have discussed the scientific basis of ETTR — interested readers can search for them. Let me briefly explain what exposing to the right means:"},{"t":"p","x":"“向右曝光”指的是通过曝光补偿的方法，在保持高光不溢出的前提下，把图像强制过曝一定程度，使得图像直方图向右移动，来减小噪点的数量，提高画质。","b":false,"xe":"\"Exposing to the right\" means using exposure compensation to deliberately overexpose the image to a degree — while keeping highlights unclipped — so the histogram shifts right, reducing the amount of noise and improving image quality."},{"t":"p","x":"首先我们来看一下直方图：","b":false,"xe":"First, let's look at the histogram:"},{"t":"img","f":"img-06","w":750,"h":458},{"t":"p","x":"直方图的横轴是图像的亮度，左边是暗区，左边边界是纯黑；越向右越亮，右边边界就是纯白。","b":false,"xe":"The histogram's horizontal axis is image brightness: shadows on the left, pure black at the left edge; brighter toward the right, with pure white at the right edge."},{"t":"p","x":"直方图的纵轴是像素记录的分布，越高则像素数量越多，上图是一个典型的正常测光、曝光的图像的直方图，左右各有一定的距离，也就是这片子里没有死黑死白的部分。","b":false,"xe":"The vertical axis is the distribution of recorded pixels — the higher, the more pixels. The image above is a typical histogram of a normally metered, normally exposed image: there's some distance from both edges, meaning no dead-black or blown-white areas."},{"t":"p","x":"噪点在直方图里的分布范围趋势图如下，我们可以看到噪点在图像里的分布主要在暗区，随着亮度的增加急剧减少，在中亮度部分左右逐渐消失，在亮区几乎没有噪点：","b":false,"xe":"The trend of noise distribution in the histogram is shown below: noise concentrates in the dark areas, drops sharply as brightness increases, mostly disappears around midtones, and is almost absent in the highlights:"},{"t":"img","f":"img-07","w":750,"h":458},{"t":"p","x":"我们把这两张片子叠加，它们的像素记录信号重叠的部分（黄色）就是噪点出现的位置和数量：","b":false,"xe":"Overlaying the two charts, the overlapping region of their recorded pixel signals (yellow) shows where noise appears and how much:"},{"t":"img","f":"img-08","w":750,"h":458},{"t":"p","x":"如果我们拍摄时用曝光补偿加曝一档得到的直方图（红线区），和之前的图像的直方图叠加：","b":false,"xe":"If we add one stop of exposure compensation when shooting, we get the histogram in red, overlaid on the previous one:"},{"t":"img","f":"img-09","w":750,"h":458},{"t":"p","x":"我们可以看到，新的图像直方图亮度增加了，但最右边并没有被右边边界切断，因此高光没有溢出而失去亮部细节，但是整个直方图向右边移动了；","b":false,"xe":"We can see the new histogram is brighter overall, but its right side isn't cut off by the right edge — the highlights aren't clipped and bright detail is preserved — while the whole histogram has moved right."},{"t":"p","x":"这就是“向右曝光”，它与噪点分布的重合部分（绿色）的面积小于正常曝光的噪点分布区（黄色），也就是说“向右曝光”后的噪点数量减少了；","b":false,"xe":"That's \"exposing to the right\": its overlap with the noise distribution (green) is smaller than the normal exposure's overlap (yellow) — meaning less noise after ETTR."},{"t":"p","x":"这就是“向右曝光”的方法减少噪点的物理原理：在记录图像信号数量同样多的情况下，向右曝光的图像噪点的数量减少了，所以记录信号与噪点信号数据的比例，即信噪比S/N ratio更大；而信噪比是衡量图像画质的一个最重要的指标，信噪比越高，画质越好。","b":false,"xe":"This is the physical principle by which ETTR reduces noise: with the same total amount of recorded image signal, the ETTR image contains less noise, so the ratio of recorded signal to noise — the signal-to-noise ratio (S/N) — is higher. S/N ratio is one of the most important measures of image quality: the higher, the better."},{"t":"p","x":"注意：在实际拍摄时向右曝光的程度每张片子都不同，过曝多少需要根据主体、场景、光线方向不同而随时调节。这是一个比较复杂的情况，限于篇幅我不展开说了，我会另外写文介绍。","b":false,"xe":"Note: in practice, the right degree of ETTR differs for every shot — how much to overexpose depends on subject, scene, and light direction, and must be adjusted on the fly. It's a complex topic beyond the scope of this article; I'll cover it separately."},{"t":"p","x":"最主要的原则是向右曝光但是要保持直方图的右边不被右边界切断，以免高光溢出失去细节。我一般在拍摄时会先对场景不同的方向试拍一下检查直方图不会溢出的前提下决定过曝多少。","b":false,"xe":"The main principle: expose to the right, but keep the histogram's right side from being cut off by the right edge, so highlights don't clip and lose detail. When shooting, I usually test-fire a few frames in different directions of the scene, check the histogram doesn't clip, and then decide how much to overexpose."},{"t":"p","x":"高ISO图像的噪点分布趋势如下图，可以看出高ISO的图像噪点比较多，但是曲线的规律是一样的——随着亮度的增加急剧减少，在中亮度区之后几乎消失，所以高ISO的片子向右曝光就更加重要：","b":false,"xe":"The noise-distribution trend for high-ISO images is shown below: high-ISO images have more noise, but the curve follows the same rule — noise drops sharply as brightness rises and nearly vanishes past midtones. So ETTR matters even more for high-ISO shots:"},{"t":"img","f":"img-10","w":750,"h":458},{"t":"p","x":"下面我们用一个实例来说明向右曝光的作用：","b":false,"xe":"Now let's use a real example to demonstrate ETTR's effect:"},{"t":"p","x":"我在很多摄影群里，这两年“向右曝光”的理论被很多人讨论过，其中有一个较为普遍的说法是随着CMOS感应器的动态范围越来越高，尤其是用尼康、索尼和中画幅相机这同样优质的感应器时，向右曝光变成没有必要甚至要低曝。","b":false,"xe":"In many photography groups these past two years, the ETTR theory has been much discussed. One fairly common claim is that as CMOS sensor dynamic range keeps improving — especially with the equally excellent sensors in Nikon, Sony, and medium-format cameras — ETTR has become unnecessary, and you can even underexpose."},{"t":"p","x":"因为相比高ISO图像，低ISO（小于ISO200）图像的暗部画质用提亮暗部亮度的方法，仍然可以获得同样的暗部细节而维持同样数量或者更少的噪点，但是可以保证高光不会溢出。","b":false,"xe":"The argument: compared with high-ISO images, a low-ISO (below ISO 200) image's shadows can be brightened to recover the same shadow detail with equal or less noise — while guaranteeing the highlights never clip."},{"t":"p","x":"我们来检验一下这个理论，下面是我用尼康新的旗舰机D6拍摄的同一场景的3张片子，第一张是不加曝光补偿的正常曝光，ISO220，第二张在相机测光的基础上（中央重量评价测光模式）低曝-0.67档，ISO160；第三张是以直方图不被右边界切断的情况下向右过曝到+2.67档（ISO1250）两张片子的曝光差是3.3档：","b":false,"xe":"Let's test this theory. Below are three frames of the same scene shot with Nikon's new flagship D6: the first at normal metered exposure, ISO 220; the second underexposed −0.67 EV from the meter reading (center-weighted metering), ISO 160; the third exposed to the right at +2.67 EV (ISO 1250) without the histogram touching the right edge. The exposure difference between them is 3.3 stops:"},{"t":"img","f":"img-11","w":750,"h":422},{"t":"img","f":"img-12","w":750,"h":422},{"t":"img","f":"img-13","w":750,"h":422},{"t":"p","x":"从三张片子的直方图我们可以看出：正常曝光、低曝光-0.67的直方图的数码信号大致集中在左边，和噪点分布区高度重合，过曝+2.67的直方图紧靠右边界但是并没有被切断即高光没有溢出，而数码信号和噪点分布区的重合度很低。我们在ACR里做一下简单的调节：","b":false,"xe":"From the three histograms we can see: the normal and −0.67 EV histograms concentrate their digital signal on the left, heavily overlapping the noise-distribution zone; the +2.67 EV histogram hugs the right edge without being cut — no highlight clipping — and barely overlaps the noise zone. Let's do a simple adjustment in ACR:"},{"t":"img","f":"img-14","w":750,"h":422},{"t":"img","f":"img-15","w":750,"h":422},{"t":"img","f":"img-16","w":750,"h":422},{"t":"p","x":"三张片子都不降噪、不锐化，高光压制到最低来保住所有细节，阴影提亮到60来增加暗区的纹理，唯一的区别是第一张曝光量增加+2.67、第二张曝光量增加+3.3与第三张曝光同步。","b":false,"xe":"None of the three is denoised or sharpened; Highlights pulled to minimum to preserve all detail; Shadows lifted to +60 to bring out dark-area texture. The only difference: exposure raised +2.67 on the first and +3.3 on the second to sync with the third."},{"t":"p","x":"然后我们打开这三张片子到Photoshop里，把鸟头放大到200%进行比较，见下图：","b":false,"xe":"Then we open all three in Photoshop and zoom the bird's head to 200% for comparison:"},{"t":"img","f":"img-17","w":750,"h":1266},{"t":"p","x":"过曝的第三张噪点很弱，极易用插件除去，鸟眼边界明显清晰，立体感强，画质极佳，即使这张片子的ISO是1250。","b":false,"xe":"The overexposed third frame has very weak noise, easily removed by a plugin; the bird's eye edges are clearly defined, dimensionality is strong, and image quality is excellent — even though this frame is ISO 1250."},{"t":"p","x":"正常曝光的第一张，鸟头噪点多，眼睛边缘不明显，暗部画质差，尤其在这个背光场景里，作为图像主体的鸟全身都是在暗区里，这个画质是很难接受的。","b":false,"xe":"The normally exposed first frame has lots of noise on the bird's head, indistinct eye edges, and poor shadow quality — especially in this backlit scene where the bird, the subject, sits entirely in the dark zone. This quality is hard to accept."},{"t":"p","x":"第二张低曝的片子鸟头充满了噪点，鸟眼的边界不可分辨，即使这张片子的ISO只有160，尤其是如果我们放大到400%如下图，我们可以看到大量的红绿蓝颜色噪点条纹，造成画质完全不可用：","b":false,"xe":"The underexposed second frame's bird head is full of noise and the eye boundary is indistinguishable — even though it's only ISO 160. Zooming to 400% as below, we see massive red-green-blue color-noise stripes, making the quality completely unusable:"},{"t":"img","f":"img-18","w":750,"h":422},{"t":"p","x":"这三张原片可以从网盘上下载，大家可以自己比较体验：","b":false,"xe":"These three RAW files can be downloaded from the cloud drive — compare them yourself:"},{"t":"p","x":"（链接: https://pan.baidu.com/s/1w2-5QmEAWas5TIqeNvZC2w 提取码: px57 ）","b":false,"xe":"(Link: https://pan.baidu.com/s/1w2-5QmEAWas5TIqeNvZC2w  Extraction code: px57)"},{"t":"p","x":"一句话总结：认为只要保持低ISO不用过曝或低曝就可以保持住画质的理论不靠谱，以直方图作为曝光标准的“向右曝光”方式是保持画质的最好方法，因为这是以感应器记录信号和噪点的物理特性决定的：噪点主要集中在暗区。","b":false,"xe":"One-sentence summary: the theory that keeping ISO low — without overexposing, or even underexposing — preserves image quality doesn't hold up. Histogram-based \"expose to the right\" is the best way to preserve quality, because it's determined by the physics of how sensors record signal and noise: noise concentrates in the dark areas."},{"t":"p","x":"“杰夫流程2.0” 第二步：RAW 文件处理——保住所有细节，包括噪点，不要降噪！","b":false,"xe":"\"Jeff Workflow 2.0\" Step 2: RAW Processing — Keep All Detail, Including Noise. Do NOT Denoise!"},{"t":"p","x":"敲黑板说重点——这一步的关键是：在ACR里不要降噪。","b":false,"xe":"Key point — the crucial thing in this step: do NOT denoise in ACR."},{"t":"p","x":"再强调第二遍：与旧版的“杰夫流程”相比这一步是最大的区别，在ACR中我们要做的最重要的一件事是：不要降噪！","b":false,"xe":"Emphasizing a second time: compared with the old Jeff Workflow, this step is the biggest difference. The single most important thing in ACR is: do NOT denoise!"},{"t":"p","x":"我们下面用这张ISO 14400的猎豹捕猎图像的实例来研究比较一下，这是原片：","b":false,"xe":"Let's study this with the ISO 14,400 cheetah-hunt image. Here's the original:"},{"t":"img","f":"img-19","w":750,"h":422},{"t":"p","x":"注意我第一步向右曝光的控制，在猎豹捕猎前等待时我拍了一些样片，确定了过曝+1.33档获得的直方图是一个完美的位置——高光信号没有被右边界切断，但是大部分的信号分布在直方图的右边，片子看上去有些发白，这是向右曝光片子的特征；","b":false,"xe":"Note my first-step ETTR control: while waiting before the hunt, I shot test frames and determined that +1.33 EV gave a perfect histogram position — highlight signals not cut off at the right edge, but most of the signal distributed on the right side. The frame looks somewhat whitish; that's the hallmark of an ETTR exposure."},{"t":"p","x":"最重要的是看着直方图，确认右边边界没被切断，相机后面的3寸预览图不能完全看清楚，所以曝光一定要以直方图为准。下面我们分析一下这张没调整过的RAW文件原片，把它放大到200%：","b":false,"xe":"Most importantly, watch the histogram and confirm the right edge isn't cut. The camera's 3-inch preview can't show this clearly, so exposure must be judged by the histogram. Now let's analyze this unadjusted RAW file, zoomed to 200%:"},{"t":"img","f":"img-20","w":750,"h":422},{"t":"p","x":"从这个局部我们得出几个结论，是我们必须在后期里要解决的：","b":false,"xe":"From this crop we draw several conclusions that must be solved in post:"},{"t":"p","x":"1.即使过曝程度控制很好，ISO14400的图像放大后的噪点数量很多，基本上都是灰色均匀的亮度噪点。","b":false,"xe":"1. Even with well-controlled overexposure, an ISO 14,400 image zoomed in shows lots of noise — mostly uniform gray luminance noise."},{"t":"p","x":"2.蓝框里我们可以看到，在右边的亮区噪点数量，要少于左边暗区的噪点，噪点分布随亮度的增加而减少。","b":false,"xe":"2. In the blue box, we can see the bright right side has less noise than the dark left side — noise decreases as brightness increases."},{"t":"p","x":"3.绿框里我们可以看到，高ISO造成的细小绒毛的边界不够清晰。","b":false,"xe":"3. In the green box, the fine fur edges caused by high ISO aren't clear enough."},{"t":"p","x":"4.红框里我们可以看到，由于豹子头左边基本在暗区，猎豹眼珠边界不是很清晰、眼白不干净与眼眶暗区的噪点混合。","b":false,"xe":"4. In the red box, since the left side of the leopard's head is basically in the dark zone, the eyeball boundary isn't clear, the whites of the eye aren't clean, and they mix with the noise in the dark eye-socket area."},{"t":"p","x":"在我使用旧版的“杰夫流程”处理这张片子时，我使用了复合式降噪的方法；我会对同一张RAW文件做3个不同程度的降噪和调节版本，把它们叠加到同一张图像里，用亮度蒙版和普通蒙版手动混合这三个不同版本图层，大约花45分钟时间，现在使用人工智能除噪软件，我只9-10分钟。","b":false,"xe":"When I processed this image with the old Jeff Workflow, I used composite denoising: I made three differently denoised and adjusted versions of the same RAW file, stacked them into one image, and manually blended the three layers with luminosity masks and regular masks — about 45 minutes of work. Now, using AI denoise software, I need only 9–10 minutes."},{"t":"p","x":"现在我们讨论一下RAW文件处理原则:","b":false,"xe":"Now let's discuss the RAW processing principles:"},{"t":"p","x":"为人工智能算法提供尽可能多的细节、纹理数据，尤其是在暗区。图像亮区的细节清晰且噪点少，暗区纹理模糊并且和噪点混合，所以把图像暗区做亮一些，能够帮助人工智能软件更准确地识别内容、区分噪点。","b":false,"xe":"Provide the AI algorithm with as much detail and texture data as possible, especially in the dark areas. Bright areas are clean and low-noise; dark areas are blurry and mixed with noise, so brightening the dark areas helps the AI software identify content and separate noise more accurately."},{"t":"p","x":"进一步对于亮度和对比度的调节，完全可以在除噪后再进一步在调图软件里进行调节。","b":false,"xe":"Further brightness and contrast adjustments can be made later in the editing software, after denoising."},{"t":"p","x":"我做的RAW文件处理如下，注意这些调节每张片子都不同，所以要注意理解为什么做这些调节更重要，而不是具体数值：","b":false,"xe":"Here's my RAW processing. Note these settings differ for every image — understanding WHY matters more than the specific numbers:"},{"t":"p","x":"1. 曝光：我们说过我们把这张片子过曝了+1.33档，但是我只降低了-0.3档是因为这只豹子的头有部分在暗区，但是我们需要它亮一点。","b":false,"xe":"1. Exposure: we overexposed this frame by +1.33 EV, but I only pulled it down −0.3 EV, because part of the leopard's head sits in the dark zone and we need it brighter."},{"t":"p","x":"2. 对比度降低：提高对比度会使得暗区更暗，减少对比度能够有效暴露暗区的像素细节，但是必须注意的一点是，对比度不能降得太低，太低会使得噪点和暗部的纹理细节像素边缘模糊，不易区分，使得降噪软件难以判断哪里是应该保持的纹理哪里是噪点。","b":false,"xe":"2. Reduce contrast: raising contrast darkens the dark areas; reducing contrast effectively reveals dark-zone pixel detail. But one caveat: don't lower it too much — too low blurs the edges between noise pixels and shadow-texture pixels, making it hard for the denoise software to tell which is texture worth keeping and which is noise."},{"t":"p","x":"做了40多张片子之后我的经验是在ACR里把对比度降到-20至-35左右效果最理想，在这个范围里，像素细节从明到暗的过渡很细腻，同时像素边缘的清晰度损失也不大。","b":false,"xe":"After processing 40+ images, my experience is that lowering contrast to around −20 to −35 in ACR works best: within this range, the bright-to-dark transition of pixel detail stays delicate, with little loss of edge definition."},{"t":"p","x":"3. 高光压制：我一般是最大限度地把高光压到最左边，起码在-80以上，因为我们会把片子做的亮一些，所以压制高光可以把亮部细节完整地保留。","b":false,"xe":"3. Highlight recovery: I usually drag Highlights all the way left, at least −80. Since we'll make the image brighter, pulling highlights down preserves bright-area detail completely."},{"t":"p","x":"4. 阴影提亮：这一步也是为了让暗部的细节更多地暴露出来。阴影提亮的程度取决于图像本身的亮度和主体（如下图猎豹身体）上最暗的地方的亮度，不能提的太亮，因为亮度噪点是灰色的，暗区提的太亮会混淆噪点和暗部纹理的区别。","b":false,"xe":"4. Shadow lifting: this step also exposes more shadow detail. How much to lift depends on the image's own brightness and the darkest point on the subject (the cheetah's body below). Don't lift too much: luminance noise is gray, and overly bright shadows blur the distinction between noise and dark texture."},{"t":"p","x":"提亮阴影程度的原则是在最黑的地方拉出纹理即可，不要太亮，既保持住明暗的光学逻辑，又能最大限度的展示所有纹理像素（见下图中红色箭头指示的猎豹身上的斑纹最黑处）。","b":false,"xe":"The principle for lifting shadows: pull texture out of the blackest areas, but not too bright — preserve the optical logic of light and dark while revealing as many texture pixels as possible (see the blackest spots of the cheetah's rosettes indicated by the red arrows below)."},{"t":"p","x":"5. 纹理：我会强化一些纹理，来平衡减少对比度造成的像素边缘不清晰的趋势，来帮助人工智能更好地识别纹理内容，这一步也要注意不能过，ACR里对焦准确的图像建议强化纹理的程度在0-25（根据我的实际操作经验）内。","b":false,"xe":"5. Texture: I strengthen Texture somewhat to counteract the softening of pixel edges caused by reduced contrast, helping the AI better recognize texture content. Don't overdo this either: for accurately focused images in ACR, I recommend 0–25 (based on my hands-on experience)."},{"t":"p","x":"【在ACR里，“纹理”、“清晰度”、“锐化”这三个调节功能都是用加局部对比度的方法来提高清晰度，“锐化”主要是像素边缘明暗对比度的强化，对于中亮度的低对比度微小纹理的强化程度不够精确，而且会增加噪点；“纹理”是中亮度细小纹理强化最精细的“高频”对比度强化，且不容易增加噪点。“清晰度”的功能介于两者之间。】","b":false,"xe":"[In ACR, \"Texture,\" \"Clarity,\" and \"Sharpening\" all increase apparent sharpness by adding local contrast. \"Sharpening\" mainly strengthens the light-dark contrast at pixel edges; it's imprecise for low-contrast micro-textures in midtones and increases noise. \"Texture\" is the most refined \"high-frequency\" contrast boost for small midtone textures and rarely adds noise. \"Clarity\" sits between the two.]"},{"t":"p","x":"6. 在ACR里，一定要把“锐化”和“减少杂色”完全关闭，数值降为0。","b":false,"xe":"6. In ACR, be sure to turn \"Sharpening\" and \"Noise Reduction\" completely off — set both to 0."},{"t":"p","x":"这一步很重要，“锐化”为0是防止更多噪点的产生，人工智能降噪软件在下一步会根据画面内容自动进行锐化。","b":false,"xe":"This step matters. Sharpening at 0 prevents generating more noise; the AI denoise software will sharpen automatically based on content in the next step."},{"t":"p","x":"“减少杂色”关闭是为了最大限度地保持图像记录的信号数据量（包括噪点），发挥人工智能算法的侦测、识别、判断能力发挥到极限。我们知道一旦我们开始在ACR里降噪，纹理细节也开始消失；要相信人工智能对于噪点识别与消除的能力。","b":false,"xe":"Turning off \"Noise Reduction\" maximizes the amount of recorded signal data (including noise), letting the AI's detection, recognition, and judgment abilities work at their limit. We know that once we start denoising in ACR, texture detail begins to disappear — trust the AI's ability to identify and remove noise."},{"t":"img","f":"img-21","w":750,"h":422},{"t":"p","x":"这几步调节做完后就可以把它打开到Photoshop里进行人工智能除噪和锐化了。","b":false,"xe":"After these adjustments, open the file into Photoshop for AI denoising and sharpening."},{"t":"p","x":"“杰夫流程2.0” 第三步：用人工智能降噪软件降噪与锐化","b":false,"xe":"\"Jeff Workflow 2.0\" Step 3: Denoise and Sharpen with AI Software"},{"t":"p","x":"敲黑板说重点——这一步的关键是：打开“自动侦测设置”，让人工智能的能力得到充分发挥。","b":false,"xe":"Key point — the crucial thing in this step: turn on \"Auto-detect settings\" so the AI's abilities are fully unleashed."},{"t":"p","x":"这一步我们就要用Topaz DeNoise AI 人工智能降噪软件来进行降噪和锐化了。","b":false,"xe":"In this step we use Topaz DeNoise AI for denoising and sharpening."},{"t":"p","x":"选用Topaz的原因是到现在为止（2020年7月），这是市场上唯一一款较为成熟的用人工智能算法降噪的软件，随着科技的日新月异，我肯定今后会有更多的这类软件出现，而我也会不断地试用，并根据效果进行测评。","b":false,"xe":"Why Topaz: as of now (July 2020), it's the only relatively mature AI-algorithm denoise software on the market. As technology advances rapidly, I'm sure more such software will appear, and I'll keep testing and reviewing them."},{"t":"p","x":"首先是在Photoshop里安装Topaz DeNoise AI软件，它也可以作为一个独立软件使用。","b":false,"xe":"First, install Topaz DeNoise AI — it can also run as a standalone application alongside Photoshop."},{"t":"p","x":"我们在ACR处理好RAW文件后打开进入Photoshop，然后1).点击“Ctrl+ J”，这样就形成一个新的图层，把这个图层重新命名为”Topaz AI 1”, 2）.然后点击顶板上的“滤镜”-“Topaz Labs”-“Topaz DeNoise AI…”：","b":false,"xe":"After processing the RAW file in ACR, we open it into Photoshop, then 1) press \"Ctrl+J\" to create a new layer and rename it \"Topaz AI 1\"; 2) click \"Filter\" → \"Topaz Labs\" → \"Topaz DeNoise AI…\" on the top bar:"},{"t":"img","f":"img-22","w":750,"h":422},{"t":"p","x":"这时Topaz DeNoise AI 的操作界面打开，第一次打开我们要做三件事：","b":false,"xe":"The Topaz DeNoise AI interface opens. The first time, we need to do three things:"},{"t":"p","x":"1. 关闭介绍弹窗；每次打开界面时这个弹窗会弹出，我们先选择“No”，再点击“关闭”键，这样弹窗以后打开时就不会弹出。","b":false,"xe":"1. Close the intro popup: it appears every time the interface opens. Choose \"No\" first, then click \"Close,\" so it won't pop up again."},{"t":"img","f":"img-23","w":750,"h":422},{"t":"p","x":"2. 设置预览模式：","b":false,"xe":"2. Set the preview mode:"},{"t":"p","x":"1）先点击“视窗”键，选择第二项“分窗预览”，这时界面中间出现一条竖线，竖线左边是除噪前，竖线右边是除噪后的预览图。","b":false,"xe":"1) Click the \"View\" button and choose the second option, \"Split View\" — a vertical line appears in the middle: denoise-before on the left, denoise-after preview on the right."},{"t":"p","x":"2）点击“放大”键，调节成100%或200%，这是检查除噪效果预览细节最直观的方式（我一般设置成200%，但每个人的习惯不一样）。这样每次界面打开后的预览形式就被固定了。","b":false,"xe":"2) Click \"Zoom\" and set 100% or 200% — the most direct way to check denoise detail (I usually use 200%, but habits differ). The preview layout is then fixed for every future launch."},{"t":"img","f":"img-24","w":750,"h":422},{"t":"p","x":"我们先来介绍一下这个操作界面：","b":false,"xe":"Let's introduce the interface first:"},{"t":"p","x":"现在的Topaz AI还没有官方的汉化版，作为专业摄影师我不能推荐那些“破解绿色汉化版”；","b":false,"xe":"Topaz AI currently has no official Chinese version, and as a professional photographer I can't recommend those \"cracked green localized versions\";"},{"t":"p","x":"不过要熟练地用这个插件我们明白下面这几个简单的调节键就够用了；其中的：","b":false,"xe":"But to use this plugin proficiently, understanding the few simple adjustment keys below is enough. Among them:"},{"t":"p","x":"4.“恢复原始细节”、5.“低光模式”6“颜色噪点降噪”这三个键基本不用，","b":false,"xe":"4. \"Recover Original Detail,\" 5. \"Low Light Mode,\" 6. \"Color Noise Reduction\" — these three are basically unused,"},{"t":"p","x":"所以我们只要掌握1，2，3这三个键的用法就可以了。","b":false,"xe":"so we only need to master keys 1, 2, and 3."},{"t":"img","f":"img-25","w":750,"h":422},{"t":"p","x":"下面我们来看看这张猎豹捕猎照片的降噪操作，进入界面后，Topaz 自动会生成一个默认的降噪结果，注意“自动侦测设置”键在默认关闭的状态；这个结果是一个均匀降噪的结果（默认降噪15，锐化15）。","b":false,"xe":"Now let's denoise the cheetah-hunt photo. Upon entering the interface, Topaz automatically generates a default denoise result — note the \"Auto-detect settings\" key is OFF by default. This default result is uniform denoising (denoise 15, sharpen 15)."},{"t":"p","x":"我们可以看见猎豹头下面的背景里仍然有可见噪点，但是豹脸上绿框里的细毛的纹理细节已经开始变弱了。当然我们可以手动改变降噪的程度，但是细节也开始丢失，这样的结果和其他的降噪软件没有太多区别，这是因为在这个过程中因为“自动侦测设置”键在默认关闭的状态，所以人工智能功能并没有完全打开：","b":false,"xe":"We can see visible noise remains in the background below the cheetah's head, while the fine-fur texture in the green box on the face has already started to weaken. Of course we could manually change the denoise strength, but detail starts to be lost — such results barely differ from other denoise software. That's because with \"Auto-detect settings\" off, the AI capability isn't fully enabled:"},{"t":"img","f":"img-26","w":750,"h":422},{"t":"p","x":"我们点击“自动侦测设置”按钮，把它打开（黄色箭头）：这时人工智能完全开放开始了侦测-识别-插值的计算然后形成了预览图如下：","b":false,"xe":"Click the \"Auto-detect settings\" button to turn it on (yellow arrow): the AI fully opens up, performing detection–recognition–interpolation calculations, and forms the preview below:"},{"t":"img","f":"img-27","w":750,"h":422},{"t":"p","x":"这个人工智能的预览版本里我们看到：","b":false,"xe":"In this AI preview we see:"},{"t":"p","x":"1）豹脸上的细绒毛的纹理被强化，锐度提高。","b":false,"xe":"1) The fine-fur texture on the leopard's face is strengthened and sharpness improved."},{"t":"p","x":"2）眼睛里眼珠边界清楚，眼白清澈，眼眶边界定义清晰，纹理干脆，这是人工智能识别和插值重新定义的结果。","b":false,"xe":"2) The eye: the eyeball boundary is clear, the whites are clean, the eye-socket edges are well defined, textures are crisp — the result of AI recognition and interpolation redefinition."},{"t":"p","x":"3）蓝框背景里的噪点右边清晰，左边部分原来的噪点被当成纹理被保留了。先不要担心，点击“应用”键，让软件重新把Topaz AI 1图层重新做一个渲染，这个过程大约是3-4分钟。","b":false,"xe":"3) In the blue-boxed background, the right side is clean, while on the left some original noise was kept as texture. Don't worry yet — click \"Apply\" and let the software re-render the Topaz AI 1 layer; this takes about 3–4 minutes."},{"t":"p","x":"我们的下一步就要解决背景里的不必要的纹理强化。","b":false,"xe":"Our next step is to deal with the unnecessary texture enhancement in the background."},{"t":"p","x":"在Photoshop里把未处理过的原片背景层再做一个拷贝，改名“Topaz 背景”，然后把这层在TopazAI里打开：","b":false,"xe":"In Photoshop, make another copy of the unprocessed background layer, rename it \"Topaz Background,\" and open this layer in Topaz AI:"},{"t":"img","f":"img-28","w":750,"h":422},{"t":"p","x":"同样的点击“自动侦测设置”键，这样又生成了与上图一样的预览图，然后我们手动把“降噪”键向右边拉，同时观察猎豹头下背景里的纹理，当纹理平滑时停住，","b":false,"xe":"Likewise click \"Auto-detect settings\" to generate the same preview as above; then manually drag the \"Remove Noise\" slider right while watching the texture in the background under the cheetah's head, and stop when the texture turns smooth."},{"t":"img","f":"img-29","w":750,"h":422},{"t":"p","x":"然后点击“应用”键，把Topaz 背景图层重新渲染。（注意当我们开始手动调解时，“自动侦测设置”就被关闭了。）","b":false,"xe":"Then click \"Apply\" to re-render the Topaz Background layer. (Note: once we start manual adjustment, \"Auto-detect settings\" turns off.)"},{"t":"p","x":"现在我们有三个图层，最上面的是Topaz AI 1图层，主体（猎豹）清晰，背景稍有不需要的纹理。","b":false,"xe":"Now we have three layers. The top is Topaz AI 1: the subject (cheetah) is sharp, but the background has some unwanted texture."},{"t":"p","x":"中间是Topaz 背景图层，背景光滑干净，主体清晰度略低。","b":false,"xe":"The middle is Topaz Background: the background is smooth and clean, but the subject is slightly less sharp."},{"t":"p","x":"最下面是未作降噪处理的原片背景层。","b":false,"xe":"The bottom is the original background layer with no denoising."},{"t":"p","x":"我们下面要做的是用蒙版把主体清晰度最高的Topaz AI 1图层的猎豹与下面Topaz 背景层里的背景部分叠加。","b":false,"xe":"Next, we'll use a mask to composite the sharpest subject from Topaz AI 1 with the background from the Topaz Background layer."},{"t":"img","f":"img-30","w":750,"h":422},{"t":"p","x":"我们首先回到最上面的Topaz AI 1图层，然后打开在Photoshop里的“目标选择”工具选择猎豹和羚羊，再换“快速选择”工具进行更精确的选择，然后点击图层工具栏下面的“添加图层蒙版”按钮：","b":false,"xe":"First go back to the top Topaz AI 1 layer. Use Photoshop's \"Select Subject\" tool to select the cheetah and the antelope, then switch to \"Quick Selection\" for a more precise selection, and click the \"Add Layer Mask\" button at the bottom of the Layers panel:"},{"t":"img","f":"img-31","w":750,"h":422},{"t":"p","x":"这样蒙版把选区以外的部分隐藏了，而只把清晰的主体保留在最上层：我们把现在看到的叠加图层再放大到100%：可以看到主体非常清晰，背景干净平滑：","b":false,"xe":"The mask hides everything outside the selection, keeping only the sharp subject on top. Zoom the composited layers to 100%: the subject is very sharp, the background clean and smooth."},{"t":"img","f":"img-32","w":750,"h":422},{"t":"p","x":"完全降噪和保持住画质后，我们现在可以调整对比度和颜色了，我们把最后完成的作品和我以前用旧版“杰夫流程”时用Dfine2.0做的版本放大到100%做一个比较。","b":false,"xe":"With full denoising and quality preserved, we can now adjust contrast and color. Let's compare the final piece at 100% with the version I made using Dfine 2.0 under the old Jeff Workflow."},{"t":"p","x":"右边的Topaz DeNoise AI 人工智能除噪后的清晰度和细节画质惊人的好——记住这是一张ISO14400的片子：","b":false,"xe":"The right side, processed by Topaz DeNoise AI, is astonishingly good in clarity and detail — remember, this is an ISO 14,400 image:"},{"t":"img","f":"img-33","w":750,"h":422},{"t":"p","x":"这也说明了另一个可能性，就是人工智能降噪软件对于高ISO图像的细节保留和画质控制，使得我们对以前画质不太满意的作品可以重做。","b":false,"xe":"This points to another possibility: AI denoising preserves so much detail and quality in high-ISO images that we can redo older works we weren't satisfied with."},{"t":"p","x":"更优秀的画质也为进一步利用人工智能的插值算法来做画幅放大提供了一个很大的可能性。对于广大摄影师来说无疑是一个福音。","b":false,"xe":"Better quality also opens the door to using AI interpolation for upscaling frame size — undoubtedly great news for photographers everywhere."},{"t":"p","x":"再次敲黑板：如果我们在ACR里降噪…","b":false,"xe":"Key point again: what if we denoise in ACR…"},{"t":"p","x":"我已经强调过两次了:用人工智能降噪最重要的是不要在ACR里降噪，那么如果我们在ACR里降噪了结果是怎样的？见下图:","b":false,"xe":"I've stressed it twice already: the most important thing with AI denoising is not to denoise in ACR. So what happens if we do? See below:"},{"t":"img","f":"img-34","w":750,"h":422},{"t":"p","x":"对于一张ISO 14400的片子，在ACR里先除掉一部分噪点（右边我用了30），和左边没有除噪的片子放大100%的比较来看，噪点（尤其是背景）似乎改善了很多，再用Topaz除噪似乎画质应该更好？","b":false,"xe":"For an ISO 14,400 image, partially denoising in ACR first (I used 30 on the right) and comparing at 100% with the un-denoised frame on the left, the noise (especially in the background) seems much improved — surely running Topaz afterward should give even better quality?"},{"t":"p","x":"错！下面是用Topaz 打开人工智能之后的片子比较：","b":false,"xe":"Wrong! Below is the comparison after opening both in Topaz with AI enabled:"},{"t":"img","f":"img-35","w":750,"h":422},{"t":"p","x":"与没有除噪的原片生成的图像相比,右边RAW文件除噪后生成的图像，红框内的背景仍有噪点，而左边原片没除过噪的原片生成图像干净平滑；绿框内的豹毛细节纹理清晰，原片除过噪的版本则相对模糊。","b":false,"xe":"Compared with the image generated from the un-denoised original, the image from the pre-denoised RAW (right) still has noise in the red-boxed background, while the left one from the untouched original is clean and smooth. In the green box, the leopard's fur texture is clear in the original-based version but relatively blurry in the pre-denoised one."},{"t":"p","x":"RAW片不除噪是为了最大限度地保持图像记录的信号数据量（包括噪点），让人工智能算法的侦测、识别、判断能力发挥到极限，用它就要相信它！预先除噪造成的像素损失会让人工智能的侦测识别能力大打折扣而失去它的优势。","b":false,"xe":"Not denoising the RAW file maximizes the recorded signal data (including noise), letting the AI's detection, recognition, and judgment work at their limit — if you use it, trust it! Pixel loss from pre-denoising greatly weakens the AI's detection ability and destroys its advantage."},{"t":"p","x":"重要的事情说第三遍：在ACR里千万不要除噪！","b":false,"xe":"Saying it a third time because it matters: do NOT denoise in ACR!"},{"t":"p","x":"对于专业摄影师来说，人工智能除噪软件对画质的保持使得我们就可以更进一步放开手脚，在最极端的环境下拍摄，而无需太顾及高ISO的限制。","b":false,"xe":"For professional photographers, AI denoising's quality preservation lets us shoot more freely in the most extreme environments, without worrying so much about high-ISO limits."},{"t":"p","x":"下面这张图片是我三天前拍摄的，这张片子使用了“杰夫流程2.0”，ISO 高达72000！但是画质非常优秀；这样的高ISO片子即使在我的拍摄生涯中也是极为罕见的。我们来看看这张片子是怎样做的：","b":false,"xe":"The image below was shot three days ago using \"Jeff Workflow 2.0\" — at ISO 72,000! Yet the quality is excellent. Such a high ISO is extremely rare even in my career. Let's see how it was done:"},{"t":"img","f":"img-36","w":750,"h":458},{"t":"p","x":"原片是这样的，ISO72000下的优秀画质还是要靠“向右曝光”，在现场我试过几张之后，确定了过曝+1.67后直方图的位置非常好，基本避开了噪点分布区：","b":false,"xe":"Here's the original. Excellent quality at ISO 72,000 still relies on \"exposing to the right\": after a few test frames on site, I settled on +1.67 EV, which put the histogram in a very good position, largely avoiding the noise-distribution zone:"},{"t":"img","f":"img-37","w":750,"h":422},{"t":"p","x":"在ACR里我做了如下调整：","b":false,"xe":"In ACR I made the following adjustments:"},{"t":"p","x":"1. 因为“向右曝光”过曝了1.67，我先把曝光减了1.7档。","b":false,"xe":"1. Because of the +1.67 ETTR overexposure, I first reduced exposure by 1.7 stops."},{"t":"p","x":"2. 减对比度来保持暗部的细节。","b":false,"xe":"2. Reduced contrast to preserve shadow detail."},{"t":"p","x":"3. 减高光来进一步保持住亮部细节。","b":false,"xe":"3. Reduced highlights to further protect bright-area detail."},{"t":"p","x":"4. 我没有提亮阴影，因为暗部的亮度已经足够","b":false,"xe":"4. I did NOT lift shadows, because the dark areas were already bright enough."},{"t":"p","x":"5. 我把白底减低，让白色变一点灰以减少片中的对比度","b":false,"xe":"5. Lowered Whites to turn white slightly gray, reducing contrast in the frame."},{"t":"p","x":"6. 强化了一点纹理来让像素边缘清晰一点","b":false,"xe":"6. Strengthened Texture a bit to keep pixel edges clear."},{"t":"p","x":"7. 锐化和降噪降到零，最大化地让所有纹理信号保持住。","b":false,"xe":"7. Dropped Sharpening and Noise Reduction to zero, maximizing preservation of all texture signals."},{"t":"img","f":"img-38","w":750,"h":422},{"t":"p","x":"打开Raw到Photoshop后，我复制一个背景原片的图层然后把图层在Topaz DeNoise AI 里打开进入界面，点击“自动侦测设置”键，生成预览图，在100%放大率来看看下图：","b":false,"xe":"After opening the RAW into Photoshop, I duplicated the background layer, opened that layer in Topaz DeNoise AI, clicked \"Auto-detect settings,\" and generated the preview. Let's look at 100% magnification:"},{"t":"img","f":"img-39","w":750,"h":422},{"t":"p","x":"所有的细节都被保持住了，人工智能在这张图里准确地定义了焦内和焦外区域，焦内主体清晰锐利，焦外平滑干净，全片无噪点，但主体细节非常棒。","b":false,"xe":"All detail is preserved. In this image the AI accurately defined in-focus and out-of-focus areas: the in-focus subject is crisp and sharp, the out-of-focus areas smooth and clean, the whole frame noise-free — with superb subject detail."},{"t":"p","x":"这样我根本不用做第二个背景版本。下面我点击“应用”键，渲染图像，回到Photoshop，放大到100%再仔细检查：","b":false,"xe":"So I didn't need a second background version at all. Then I clicked \"Apply\" to render, returned to Photoshop, and zoomed to 100% for a careful check:"},{"t":"img","f":"img-40","w":750,"h":422},{"t":"p","x":"我们可以看到ISO72000无噪点，而鸟头暗区纹理，小龙虾的细节，幼鸟柔细的羽毛所有主体细节都非常清晰，而焦外背景则非常柔滑。这张片子人工智能对于图片内容的分析和识别处理近乎完美。","b":false,"xe":"We can see: ISO 72,000 with no noise, while the dark-area texture on the bird's head, the crayfish's details, and the chick's soft feathers are all very clear, and the out-of-focus background is silky smooth. The AI's analysis and recognition of this image's content is nearly perfect."},{"t":"p","x":"我们接下来把画面剪裁，最后图像的长边仅仅3188像素：","b":false,"xe":"Next we crop the frame — the final image's long edge is only 3,188 pixels:"},{"t":"img","f":"img-41","w":750,"h":422},{"t":"p","x":"我们再进一步调节局部亮度、对比度、颜色，这就是最后的ISO72000的作品：","b":false,"xe":"We then fine-tune local brightness, contrast, and color — and here is the final ISO 72,000 work:"},{"t":"p","x":"“杰夫流程2.0” 第四步：画质控制最后一步，缩图与锐化","b":false,"xe":"\"Jeff Workflow 2.0\" Step 4: Final Quality Control — Resizing and Sharpening"},{"t":"p","x":"“杰夫流程2.0”的锐化基本与旧版没有多少改变。","b":false,"xe":"Sharpening in \"Jeff Workflow 2.0\" is largely unchanged from the old version."},{"t":"p","x":"我们拍摄的数码照片的用途无非两种：一，在网络上与别人交流，如发帖子，参加摄影比赛等。","b":false,"xe":"Our digital photos have exactly two uses: first, sharing online — posting, entering photo contests, and so on."},{"t":"p","x":"二，打印输出，到不同尺寸成为印刷品，如在杂志、报纸上发表，做展览、印招贴画，或打印成照片挂在墙上。","b":false,"xe":"Second, print output at various sizes — publication in magazines and newspapers, exhibitions, posters, or prints for the wall."},{"t":"p","x":"一.网络输出：一般我们的片子99%是用于在网络上分享，或者参加数码摄影比赛，一般来说有两个共性，一是尺寸不能过大，最高到2000像素的jpeg格式。","b":false,"xe":"1. Web output: 99% of our images go to online sharing or digital photo contests. Two common traits: the size can't be too large — JPEG at up to 2,000 pixels;"},{"t":"p","x":"二是网络通用的色彩空间是sRGB，所以网络发片想要把照片的色彩空间转成sRGB。","b":false,"xe":"and the web's standard color space is sRGB, so photos for the web should be converted to sRGB."},{"t":"p","x":"网络发片的缩图过程会造成图像锐度的损失，而过度锐化又会产生噪点，所以过程的控制特别重要。以下是我现在使用的网络发图的缩图锐化的方法：","b":false,"xe":"Resizing for the web costs sharpness, while over-sharpening creates noise — so process control matters greatly. Here's the resize-and-sharpen method I currently use for web posting:"},{"t":"img","f":"img-42","w":750,"h":422},{"t":"p","x":"以猎豹片为例 ，比如我们要把这张片子缩成1600px（像素）在网络上发表","b":false,"xe":"Take the cheetah image: say we want to resize it to 1,600px for online publication."},{"t":"p","x":"1. 首先是把我照片的色彩空间从工作空间的ADOBE RGB，使用“转换为配置文件”功能，转为sRGB：","b":false,"xe":"1. First, use \"Convert to Profile\" to convert the photo's color space from the working space Adobe RGB to sRGB:"},{"t":"img","f":"img-43","w":750,"h":422},{"t":"p","x":"2. 把图像尺寸缩为最终尺寸（1600px像素）的两倍（3200px像素）：","b":false,"xe":"2. Resize the image to twice the final size (3,200px):"},{"t":"img","f":"img-44","w":750,"h":422},{"t":"p","x":"3. 复制一个图层, 然后把图层混合模式从“正常”改为”明度(Luminosity)”：","b":false,"xe":"3. Duplicate a layer, then change the layer blend mode from \"Normal\" to \"Luminosity\":"},{"t":"img","f":"img-45","w":750,"h":422},{"t":"p","x":"这一步是把上面的复制图层和原图图层的相互作用方式变为根据亮度（明度），也就是说我们等一下锐化了上面的复制层后，画面越亮的地方锐化幅度越大，而越暗的地方锐化幅度越小，这样就有效的防止了额外噪点的产生，因为噪点最容易产生在暗部。","b":false,"xe":"This makes the duplicated layer interact with the original based on luminosity — after we sharpen the duplicated layer, brighter areas get sharpened more and darker areas less, effectively preventing extra noise, since noise most easily appears in dark areas."},{"t":"p","x":"4. 对上层复制层进行智能锐化 设置-数量（Amount）：50%；半径（Radius）：0.3px(像素) ；减少杂色（Noise）: 0--（因为我们已经降过噪了）。重复锐化共3次。","b":false,"xe":"4. Apply Smart Sharpen to the duplicated layer with: Amount 50%; Radius 0.3px; Reduce Noise 0 — because we've already denoised. Repeat the sharpening 3 times in total."},{"t":"img","f":"img-46","w":750,"h":422},{"t":"p","x":"5. 缩图至最终尺寸（1600px像素）","b":false,"xe":"5. Resize to the final dimensions (1,600px)."},{"t":"img","f":"img-47","w":750,"h":422},{"t":"p","x":"6. 对上层复制层再次进行智能锐化，这次的设置-数量（Amount）：50%；半径（Radius）: 0.2px(像素)；减少杂色（Noise）: 0。","b":false,"xe":"6. Apply Smart Sharpen to the duplicated layer again, with: Amount 50%; Radius 0.2px; Reduce Noise 0."},{"t":"p","x":"重复锐化共三次。这时的画面锐度会稍微过一点。","b":false,"xe":"Repeat the sharpening 3 times in total. The image will now look slightly over-sharpened."},{"t":"img","f":"img-48","w":750,"h":422},{"t":"p","x":"7. 最后调节上面锐化图层的不透明度，同时检查图像直到锐度和细节合适。不透明度的调节取决于","b":false,"xe":"7. Finally, adjust the opacity of the sharpened layer while checking the image until sharpness and detail look right. The opacity depends on:"},{"t":"p","x":"1）：原片的锐度，原片锐则不透明度要低一些。2）：Photoshop 的版本。我的经验是如果用PhotoshopCS6, 或Photoshop CC2016之前的版本，不透明度约在18-35之间，如果是Photoshop CC2018,2020版则在5-15之间效果最好。","b":false,"xe":"1) The original's sharpness — the sharper the original, the lower the opacity. 2) The Photoshop version. In my experience, with Photoshop CS6 or versions before CC 2016, opacity around 18–35 works; with Photoshop CC 2018/2020, 5–15 gives the best results."},{"t":"p","x":"锐度满意后我们就可以合并图层存储图像了。","b":false,"xe":"Once satisfied with the sharpness, we can flatten the layers and save the image."},{"t":"img","f":"img-49","w":750,"h":422},{"t":"p","x":"二，打印输出：打印输出往往由打印的杂志社、报纸，商业单位的要求决定，他们会要求你交图的尺寸，色彩空间，有些要求你锐化，有些则不要求而由他们的技师来最后调整。","b":false,"xe":"2. Print output: printing is usually dictated by the magazine, newspaper, or commercial client's requirements — they'll specify the size and color space of the files you deliver; some ask you to sharpen, others don't and have their technicians do the final adjustment."},{"t":"p","x":"如果你自己做展览打印，又会根据你的打印尺寸，使用的纸张，打印机的厂家，型号而有所不同。在这里无法一一赘述。但是如果你要自己做锐化时一定要记住三个原则：","b":false,"xe":"If you print for your own exhibition, settings vary with print size, paper, printer brand, and model — too much to cover here. But if you sharpen yourself, remember three principles:"},{"t":"p","x":"1，一定要建立一个复制图层，把混合模式改成明度，以最大限度的控制额外噪点的产生。","b":false,"xe":"1. Always work on a duplicated layer with the blend mode set to Luminosity, to minimize extra noise."},{"t":"p","x":"2，一定要分两步逐级锐化，数量还是50%，少量多次的效果好过一步到位，半径的设置根据尺寸的不同多试几次，尺寸减小，则半径相应减少一些，原则还是防止过锐和额外噪点的产生。","b":false,"xe":"2. Always sharpen progressively in two steps, still at 50% Amount — several light passes beat one heavy one. Try different Radius settings for different sizes; the smaller the size, the smaller the radius. The principle remains: prevent over-sharpening and extra noise."},{"t":"p","x":"3，Topaz Denoise AI 的“自动侦测设置”打开后，人工智能功能本身在渲染时会进行锐化，因此与传统的TIFF片子相比，它生成的图像本身的锐度会高，在调节锐化层的不透明度时要注意观察防止图像过锐。","b":false,"xe":"3. With Topaz DeNoise AI's \"Auto-detect settings\" on, the AI itself sharpens during rendering, so its output is inherently sharper than a traditional TIFF file — watch the image when adjusting the sharpening layer's opacity to avoid over-sharpening."},{"t":"p","x":"“杰夫流程2.0——高ISO图像人工智能降噪与画质控制流程”到这里就讲完了。","b":false,"xe":"That concludes \"Jeff Workflow 2.0 — AI Denoising and Quality Control for High-ISO Images.\""},{"t":"p","x":"人工智能软件在摄影领域从相机硬件到后期软件的越来越快、越来越广的应用是最新的技术趋势。","b":false,"xe":"AI software's ever-faster, ever-wider application in photography — from camera hardware to post-processing software — is the newest technological trend."},{"t":"p","x":"未来的人工智能技术在摄影所有方面的应用，会以极高的速度取代传统的技术技巧。对于摄影师来说，想要站在时代的前缘地带，我们就不应该去排斥而是要带着更大的热情去拥抱这个时代的到来。","b":false,"xe":"Future AI technology will replace traditional techniques across all aspects of photography at tremendous speed. For photographers who want to stand at the forefront of the times, we shouldn't reject this arrival — we should embrace it with greater enthusiasm."},{"t":"p","x":"人工智能的软件还在完善中，但是现在它已经可以最大限度地帮我们摄影师保持住作品的画质，在低光高ISO场景拍摄的作品的优秀质量带给了我们更多的创作空间和灵活性。","b":false,"xe":"AI software is still being perfected, but it can already help us photographers preserve our work's quality to the maximum. The excellent quality achievable in low-light, high-ISO shooting gives us more creative space and flexibility."},{"t":"p","x":"我想最后再强调一下“杰夫流程2.0”的三个最重要的点，再给大家两个建议:","b":false,"xe":"Finally, let me re-emphasize the three most important points of \"Jeff Workflow 2.0,\" plus two suggestions:"},{"t":"p","x":"1.“向右曝光”更加重要，暗部能否曝亮是关乎到人工智能软件侦测识别的准确度，从而获得高画质最重要的前提。","b":false,"xe":"1. \"Expose to the right\" matters even more — whether the shadows can be brightened is the most important prerequisite for the AI's detection accuracy and thus for high image quality."},{"t":"p","x":"2. RAW文件不要降噪不要锐化，交给人工智能软件去做。","b":false,"xe":"2. Don't denoise or sharpen the RAW file — leave that to the AI software."},{"t":"p","x":"3.在使用Topaz DeNoise AI 时一定要打开“自动侦测设置”，充分发挥人工智能算法的优势。","b":false,"xe":"3. When using Topaz DeNoise AI, be sure to turn on \"Auto-detect settings\" to fully leverage the AI algorithm."},{"t":"p","x":"两个建议：","b":false,"xe":"Two suggestions:"},{"t":"p","x":"1. 不断更新人工智能软件，它的算法会越来越精确，越来越先进；我也会不断地尝试新的软件和大家分享体验。","b":false,"xe":"1. Keep your AI software updated — the algorithms will get ever more precise and advanced. I'll also keep trying new software and sharing my experiences."},{"t":"p","x":"2. 如果人工智能软件没达到你的要求，或者它就是不断“犯二”做出种种奇怪的事情，你可以回去再用旧版的“杰夫流程”去解决高ISO的噪点问题。","b":false,"xe":"2. If the AI software doesn't meet your needs, or it keeps acting up and doing strange things, you can always go back to the old \"Jeff Workflow\" to solve high-ISO noise problems."},{"t":"p","x":"最后我还是用这句话来说明为什么我要总结并公开分享“杰夫流程2.0”：","b":false,"xe":"I'll close with the same line as before, to explain why I summarized and publicly shared \"Jeff Workflow 2.0\":"},{"t":"p","x":"“愿此文帮到天下摄友”","b":false,"xe":"\"May this article help photographers everywhere.\""},{"t":"p","x":"谢谢关注！","b":false,"xe":"Thanks for reading!"},{"t":"p","x":"杰夫 写于多伦多，2020.7.","b":false,"xe":"Jeff, written in Toronto, July 2020."},{"t":"img","f":"img-50","w":750,"h":500},{"t":"p","x":"作者：杰夫","b":false,"xe":"Author: Jeff"},{"t":"h","x":"---《野性之美—野生动物摄影手记》作者","xe":"— Author of \"The Beauty of Wildness: A Wildlife Photography Journal\""},{"t":"p","x":"--- 加拿大专业摄影师协会PPOC（Professional Photographers Of Canada）认证的职业野生动物与自然风光摄影师","b":false,"xe":"— Professional wildlife and landscape photographer certified by the Professional Photographers of Canada (PPOC)"},{"t":"p","x":"---英国索伦特，水星，卡特斯三家新闻图片社签约摄影师","b":false,"xe":"— Contract photographer for three UK news agencies: Solent, Mercury, and Caters"},{"t":"p","x":"---肯尼亚Game Watcher Safaris 常驻摄影师","b":false,"xe":"— Resident photographer at Game Watcher Safaris, Kenya"},{"t":"p","x":"---肯尼亚Lentorre Lodge 常驻摄影师","b":false,"xe":"— Resident photographer at Lentorre Lodge, Kenya"},{"t":"p","x":"---尼康世界摄影大赛”Nikon Photo Contest\"评委","b":false,"xe":"— Judge for the Nikon Photo Contest"},{"t":"p","x":"---世界最佳自然摄影比赛“Nature's Best Photography Competition” 非洲区评委","b":false,"xe":"— Africa-region judge for Nature's Best Photography Competition"},{"t":"p","x":"---加拿大摄影艺术协会持证裁判","b":false,"xe":"— Certified judge of the Canadian Association for Photographic Art"},{"t":"p","x":"---2015,2017多伦多国际摄影节裁判委员会主席","b":false,"xe":"— Chair of the Jury Committee, Toronto International Photography Festival, 2015 & 2017"},{"t":"p","x":"---中国鸟网国际推广总监","b":false,"xe":"— International Promotion Director of China Bird Net"},{"t":"p","x":"---加拿大加华杰作摄影协会主席","b":false,"xe":"— President of the Canada Chinese Masterpiece Photography Association"},{"t":"p","x":"---上海锋锐影像文化传播公司首席策展人兼艺术总监","b":false,"xe":"— Chief Curator and Art Director of Shanghai Fengrui Image Culture Communication Co."},{"t":"img","f":"img-51","w":500,"h":500}],"titleEn":"Great Shots at ISO 70,000? This Noise Reduction Tech Is Incredible"}