How to judge a camera’s imaging

[If you’re wondering where yesterday’s post on the base ISO of the a7RII went, I pulled it. Bill Claff, Jack Hogan, and Iliah Borg helped me see that my model for an Aptina DR-Pix camera was grossly oversimplified, resulting in erroneous conclusions. I hope to have a better version to post before too long.]

A few days ago, I presented my findings on the below-base ISO settings on the Sony a7RII. I also posted on DPR the test results and my recommendation against using the below-base ISOs, which I called “fake ISOs” because they don’t actually reduce sensor sensitivity as measured by the raw file values. The thread rapidly reached the maximum number of posts and was therefore locked.

A lot of the posts amounted to an argument between people who, as I did, saw no difference in a7RII images exposed at ISO 100 and those exposed at ISO 50, save a difference in the preview image, and those who saw a big difference. It seemed like the first group was, as I was, looking at the data in the raw files to reach their conclusions. It seemed like the second group was looking at the images as displayed in their favorite raw converter to make their judgements. Both Lightroom (Lr) and its under-the skins twin Adobe Camera Raw (ACR) automatically reduce the brightness of ISO 50 a7RII images by a stop from those of ISO 100 a7RII images, the people who used those programs to draw their conclusions saw differences between the results of the two ISO settings.

It seemed to me that the two groups kept talking past each other because they were using different standards. I have seen other threads on DPR where this seemed to happen.

So I took a poll on DPR, asking how many people ever looked at the raw values in their files. Turns out, about three quarters of the respondents never look at the raw files themselves. There are two populations out there, and the population that never looks at raw values or images outnumbers those that do. As one of the responders to the poll pointed out, the actual proportion may skew even more dramatically to the never-look-at-raw group, since the slightly techie title of the poll may have prevented members of that group from participating.

What’s this mean to Internet discussions of cameras? I think it might explain many frustrating and inconclusive conversations about camera imaging characteristics.

Consider this generalization of the fake ISO discussion.

  • Someone makes a claim about something the camera does based on looking at the raw files.
  • Someone else challenges that claim based upon looking at the converted results in one or more raw developers.
  • Discord and discontent ensue.

Or it could go the other way:

  • Someone makes a claim about something the camera does based on looking at the raw files as developed by, say, Lr.
  • Someone else challenges that claim based upon looking at the raw files.
  • Sparks fly.

One of the insidious things about these discussions is usually that the different means of reaching conclusions is not explicit and clear. Even if it were, the people who are looking at the raw files might say their way is a better way to evaluate a camera, because the raw files show what the camera is doing, while the developed files show what the combination of the camera and the raw developer is doing; you can’t tell what the camera is doing by looking only at what the developer produced. The people basing their conclusions of the developed files might say that their way is better because no one can use an undeveloped raw file for a normal photographic purpose, and the best way to judge a camera’s suitability for a particular task is to perform that task and look at the results.

One of the trends of the past decade or so, led by Adobe, but followed by many, is for raw developers to become more capable and more opaque. Lr does more and more to — and/or for – your images than it used to, and it’s harder and harder to figure out what exactly it’s doing. Because of that, I expect this problem to get worse in the future.

I don’t see any easy resolution of this schism, but I do think that recognizing that there are two camps here, both with some justification for their perspective, may be the first step toward bridging the gap.

Fake ISOs, ETTR, & WYSIWYG

I posted my test results and conclusions on the a7RII’s fake ISO settings on the DPR alpha 7 forum. To put it mildly, a lively discussion ensued.

The SNR was not particularly high, but even so, as is usual in Internet discussions, I learned a couple of things. In my origina post, I took the same stance I took on this blog: don’t even think about using the fake ISOs. But, based on the discussion, I am reconsidering, at least for sophisticated photographers and two corner cases. One is a possible symbiotic relationship between ETTR and  fake ISOs.

The particular fake ISOs I’m talking about are ISO 50 on the a7RII, and presumably on the other alpha 7 cameras, and L 0.3, L.07, and L 1.0 on the Nikon D810. It doesn’t apply to ISO 80 and ISO 64 on the a7RII, because those ISOs actually reduce the dynamic range from that achievable at base ISO.

Let’s say you are using the spot metering technique for ETTR. You find a significant highlight, meter it, and, to use Zone System terms, place it on Zone VIII. You note that, since the subject was fairly low contrast, that your intended exposure is a stop over what the gray card exposure would be for the subject at base ISO. The normal ETTR technique would be to set the exposure at base ISO the way your highlight on Zone VIII meter reading said to, expose at base ISO, and pull the image a stop in post.

However, there is another alternative. You could use the same exposure and set the camera to ISO 50 in the case of the a7RII, or ISO L 1.0 in the case of the D810. the values in your raw file would be the same. However, when you opened the image in Lightroom, the 1 stop pull would have already been performed for you. Certainly not a compelling advantage, but it may be worth something to some people.

If you use the in-camera histogram, the benefits are less clear, since you trade in-the-field complexity for post-processing simplicity. You’d set the camera at base ISO, take a picture and examine the three-channel histogram (tweaked or not). When you found an exposure to your liking, and you noticed that the exposure was a stop over the gray card exposure, you could set the camera to ISO 50 or L 1.0 and make your real exposures. If you’re using he a7RII, and adjusting the exposure with the exposure compensation wheel, then noticing that you’re giving the picture a stop more than the meter reading is easy, but you have to remember to dial the exposure compensation back a stop if you switch to base ISO.

My own personal opinion is that these games aren’t worth the candle, but you may have a different opinion.

Another issue that was brought up on DPR is that the fake ISOs are useful to preserve what you see is what you get (WYSIWYG) in situations where you are intentionally overexposing  relative to the gray card exposure. An example is balancing natural light and flash.

I think this has merit in some situations as well, although I am not a big fan of using the camera’s EVF or review function to judge image quality in general.. The one place where I do find it useful is in flash balancing. In the old days, I used to use Polaroid back for this. Now, I use the LCD display. Sure, the colors are wrong, and the image is too punchy to show you what you can do in the shadows in Lr, but you can get the flash ratios dialed in well.

Now let’s say that you’ve picked a wide aperture to get the background ou of focus just the right amount, and you can’t crank up your shutter speed far enough to get a proper exposure at base ISO without getting above the camera’s maximum synch speed. You’ve also noticed that you can give the exposure a stop more without clipping the highlights. If you followed my earlier suggestions, you’d set the camera to base ISO, the shutter speed to max synch speed, overexpose by a stop, and pull the image a stop in post.

But you could also set the camera to a fake ISO setting a stop down from base ISO and use the same exposure. Then your preview image would not reflect the overexposure, and you could more easily judge the flash ratios. You’d also have the advantage that the one-stop pull in post would happen automatically.

I’m not going to use either of these techniques myself, but in the interest of explaining the limitations of my precious recommendations, I felt it desirable to write this post.

 

 

Nikon D810 fake ISOs

Yesterday I talked about the way that the ISOs below base ISO are implemented in the Sony a7RII. Today I’ll turn my attention to the Nikon D810, using the same methodology. You might want to look at yesterday’s post if you haven’t seen it already.

Here are the statistics for a series of images exposed with aperture priority at ISOs 64, 50, 40, and 32. (To their credit, Nikon doesn’t actually refer to the below-base ISO settings that way, but rather as L 0.3, L 0.7, and L 1.0.).

nikon64

nikon 50

nikon 40

Nikon 32

You can see that all of the fake ISOs are overexposed with respect to the base ISO image. However, the pattern is different than for the a7RII. The first “pulled” ISO is a third stop overexposed. The second “pulled” ISO is two=thirds stop overexposed. The third, and ultimate, “pulled” ISO is a full stop stop overexposed. In the a7RII all the “pulled” ISO settings resulted in a one-stop overexposure.

Looking at the histograms of the ISO 64 and ISO 32 image, you can see how the use of the fake ISO results in clipping:

ISO 64 hist

ISO 32 hist

But, as with the D810, bringing the images into Lightroom hides the overexposure. Here are all four pictures in the Gallery:

Lr d810 fake iso

In the a7RII, there was a loss of engineering dynamic range in going from ISO 100 to ISO 80, recovering slightly  as ISO 64, and equaling the ISO 100 EDR at IDO 50.

Here’s our white frame/black frame spreadsheet for the D810. The white frames are the first four rows, an the black frames at the last four rows.

D810 light dark stats

You can see that the clipping points are the same for all four ISOs. So are the dark field standard deviations. That’s a little better than the a7RII, which loses engineering dynamic range from the ISO 100 levels at both ISO 80 and ISO 64.

But the take home is the same: don’t use the fake ISOs on the D810. If you must overexpose, at least do it with the manual controls or the exposure compensation adjustment, so you can see what you’re doing with the preview image and the in-camera histogram.

 

Sony a7RII fake ISOs

From the mailbag:

I told [name] that the ISO values below 100 on the A7R2 are “fake” and that the dynamic range measurements you provided for those ISO values are spurious. I tried to find where you discussed that and was not able to.

[name] was asking for a link. Do you have anywhere on your blog where you discuss ISO 50 and how the dynamic range there is actually 1 stop lower than ISO 100?

I realized that I’d never addressed the fake ISO issue in any detail, contenting myself by referring to it with ill-concealed, but unexplained, disdain. I also realized that I’d never looked at the way the alpha seven cameras implemented the fake ISOs.

If you’re a camera manufacturer, there are at least two approaches possible for providing ISOs below the base ISO.

The zeroth is for the engineers to say to the product managers: “No, we’re not going to do fake ISOs on this camera. If you want it done, get somebody else to design this sucker.” That’s what I wish would happen with every camera.

The first is manipulating the raw data so that the images exposed at the fake ISOs are darker than they would otherwise be. This lowers the saturation point of the resultant raw file, and causes the dynamic range calculations to be wrong if this is not compensated for.

The second is to set the ISO to a higher value than indicated by the knob, record the result in the raw file, but change the processing of the JPEG preview image to make it darker than it would be if it were a straight rendition of the raw file. This is what the a7RII does.

You can prove the to yourself by making a set of raw and JPEG images of the same subject at ISOs 100, 80, 64, and 50, opening up the lens a third of a stop each time. If you look at the JPEG images, you’ll see that they look like the all have the same brightness. However, if you look at the raw images in RawDigger, you’ll see that the ISO 80, 64, and 50 images are all almost a stop overexposed from the ISO 100 image.

iso 100 statsa

iso 80 statsa

iso 64 statsa

iso 50 statsa

If you look at the log histogram of the ISO 100 image, you’ll see that all raw channels are short of clipping.

iso 100 hist

If you look at the log histogram of the ISO 50 image, you can see that the green channel is clipped.

iso 50 hist

But here’s something passing strange. When you import the four images into Lightroom, they look like they are all exposed more-or-less to the same brightness:

4exp Lr

Apparently, the camera passes enough information to the raw developer that, in the absence of a program like RawDigger, it’s not possible to see that the ISOs other than 100 are overexposed. The effect is that the photographer who is foolhardy enough to use the ISOs below 100 will suffer a stop loss in headroom — the avoidance of which was presumably the reason for venturing into the land of sub-base ISOs in the first place. To make the plot thicker, Lr and Sony conspire to hide the loss of headroom.

Now let’s look at two things and how they behave at ISOs of 100, 80, 64, and 50: dark-field noise and saturation level. I’ll walk you through how I make the measurements in sufficient detail that anyone can follow along in the comfort of their home or office, and can do so without resorting to charts, special lighting, or any software beyond RawDigger and the spreadsheet of your choice.

Ready?

Take an a7RII, or any other Sony alpha 7 camera in hand. Set the ISO to 100, the shutter speed to 1/125, the lens aperture to as narrow as it will go, and make an exposure of the back of the lens cap. Make three more exposures at ISOs of 80, 64, and 50.

Now, take the lens cap off, open the lens up all the way, set the ISO back to 100, and the shutter to some speed that will give a five or six stop overexposure. Make three more exposures at ISOs of 80, 64, and 50.

Upload the images to your computer, and put them in a folder of their own. Take the first one and drop it on the RawDigger icon on your desktop. When RawDigger opens, make sure that black point subtraction is disabled.

rd prefs

Now go the the Set Selection by Numbers Dialog and set up a 400×400 central sample:

selection

Convert the selection to a sample. <Ctrl>M will get you the dialog on a PC.

Open the samples window:

samples

Click on Save Samples. Accept the default file name.

Click the Append File box. Hit <Ctrl> right arrow to get the next file, and then Save Samples to append the data to the CSV file. Keep doing that until you have saved the data from all 8 files.

Close RawDigger. Double-click on the CSV file to open it in a spreadsheet program. For dark-field noise, the data you care about are the standard deviations; they’re in the columns Rdev, Gdev, and Bdev. For saturation, the data you care about are in the columns Rmax, Gmax, and Bmax. If you delete the other columns, your spreadsheet will look like this:

ss

You can see that the ISO 50 dark field standard deviations in the fourth row are the same as the ISO 100 standard deviations in the first row. You can also see that the ISO 80 standard deviations in the second row and the ISO 64 standard deviations in the third row are worse. In fact, if you’d have done dark field exposures at ISO 125 and 160, you would have found that the ISO 80 and ISO 160 standard deviations are the same, and that the ISO 64 and ISO 125 standard deviations are the same.

If you look at the overexposed images in the last four rows, you can see that the saturation values are the same no matter where you set the ISO knob.

So, the graphs I presented earlier for the engineering dynamic range of the fake ISOs were indeed accurate; the raw file saturation level is not any lower at those ISOs. However, if you want to take advantage of the lower ISOs by taking a properly exposed, just-short-of-clipping image at base ISO, dial down the ISO and give more exposure, you’re going to clip the highlights.

So the fake ISOs are, for raw shooters, a complete shuck, and a misleading one at that. If you’re using the in-camera histogram to get ETTR exposure, you’re in danger of blowing the highlights at less than base ISO, since the in-camera histogram is derived from the JPEG preview image, which is artificially darkened at ISO settings below the base ISO.

The take home lesson from many of my photographic tests is sometimes complicated and difficult to explain. This one is not: if you shoot raw, don’t ever use the fake ISO settings.

Real-world Sony a7RII autofocus

In yesterday’s post, I talked about a situation that the Sony a7RII autofocus didn’t handle very well. Actually, as I said then, I think the camera’s autofocus is probably doing fairly well, but the combination of a slowish autofocusing lens (compared to some moderately rapid to breathtakingly rapid Nikon lenses) and EVF lag and blackout made the experience unpleasant and unproductive compared to using a good SLR in the same circumstance.

Today I’d like to talk more generally on my experiences with the autofocus systems on the a7RII, after five 1000+ shot days using mostly the Zony 35mm/2.8 FE, the 90mm/2.8 OSS FE Sony macro lens, and the 70-200mm/4 Sony OSS FE zoom.

I’ll sprinkle some examples of the camera doing well with difficult subject matter throughout this post.

70-200, f?5.6 @ 1/500, ISO 100, plus 1/3 stop in post.

70-200, f/5.6 @ 1/500, ISO 100, plus 1/3 stop in post.

In a word, the autofocus is remarkable. There is very little hunting in most circumstances. You don’t have to calibrate it, because Sony uses contrast detection to trim up the focus that it got from the phase detection pixels. This gives you the best of both worlds. It’s fast, and it’s precise.

90mm macro f/2.8 @ 1/320, ISO 6400, + 2 stops (!) in post

90mm macro f/2.8 @ 1/320, ISO 6400, + 2 stops (!) in post

How precise? I did a laboratory test a week or so ago, and it did as well as I could focus manually, except at the widest f-stops. But lab tests don’t tell you how something as complex and multifaceted as an autofocus system will do in the field, and that one was set up to stack the deck in favor of autofocus anyway.

90mm macro f/2.8 @ 1/320, ISO 800, + 1 stop in post

90mm macro f/2.8 @ 1/320, ISO 800, + 1 stop in post

Here’s the bottom line. I am continually amazed at the focus accuracy I see when I zoom in in Lightroom to check focus, even in difficult situations. When autofocus fails on the a7RII, and it doesn’t fail often,  it is generally a fairly gross error, one that you can see on a 4K monitor with the image filling the frame.

Swing was moving fast, one handed camera, f/5.6 @ 1/500, ISO 100, Plus a stop in post, +71 shadow move

Swing was moving fast, one handed camera, Zony 35, f/5.6 @ 1/500, ISO 100, Plus a stop in post, +71 shadow move

In fact, where zooming in to check focus used to be something that I did all the time when culling images, now I’m starting to check a few at the start of a sequence, and, if they’re fine, I pick the images I want to carry to the next step by their content, knowing that they’re going to be sharp.

That is a huge change for me.

 

Subject on a swing. One handing the camera with the Zony 35, wide open at 1/500, ISO 100, pushed 2/3 stop in post, lus a +61 shadow move.

Subject on a swing. One handing the camera with the Zony 35, wide open at 1/500, ISO 100, pushed 2/3 stop in post, plus a +61 shadow move.

It’s going to take me a while to understand all the autofocus options, and longer still to learn which is best in which situations, and longer still to learn to trust the AF, but color me impressed.

 

a7RII operation with highly active subjects

I have tested the accuracy of the Sony a7RII autofocus for static subjects; the results were impressive. I have used the camera for many actual photographic people shots. On the whole, the autofocus systems (I say systems, because there are a dizzying number of modes and sub-modes) have performed their task with remarkable flexibility and precision. The combination of phase detection (to get close) and contrast detection (to trim up any residual errors) is powerful in concept, and generally well-executed in practice. Of course, there are always things I’d like to see changed in today’s cameras and I’ll get to my suggestions for improvement in a future post.

Rather than post an endless and unenlightening procession of “look, it focused that perfectly” images, today I’m going to concentrate on what happens when you take the camera out of its comfort zone. There are those who will call me a Cassandra. While I understand that paying more attention to the negative aspects rather than the positive ones does not give a balanced view of the camera, today’s cameras all do so many things so well that a review that uniformly emphasized what’s good and what’s bad would be very long, much the same from camera to camera, and not provide much content that would help people get the most out of the camera in question in proportion to the length of the review.

If a camera does some thing just fine, the photographer doesn’t have to think about that thing at all. If a camera has deficiencies — and they all do — the photographer is benefited by knowing about them, so that she can avoid the situations which bring the deficiencies into play, or mitigate them with workarounds.

Last month, I took a few images of my oldest granddaughter swimming butterfly. I used the Nikon D810 and the Nikon 80-400mm f/4.5-5.6 ED VR AF-S Nikkor (the name doth flow trippingly off the tongue, does it not?), with the lens set to 400mm, and the aperture set wide open, which, at that focal length, is f/5.6. I set the drive to continuous low, and the autofocus mode to continuous.

Here’s what resulted:

_8100064-2

_8100068-2

 

Yesterday, with Sony a7RII and Sony 70-200mm f//4 OSS FE in hand, I asked me granddaughter to swim some ‘fly. I set the aperture to f/5.6 and the focal length to 200mm. I experimented with letting the camera find the subject, with flexible spot mode, with single shot, with continuous low and continuous high, with pretty much everything I could think of. I never got a good shot.

I think I know a few reasons why.

When a person swims butterfly coming towards you, there are moments when she is under the water, and moments when she is out of the water. this confuses continuous autofocus, which works by predicting the subject’s rate of closure with the camera. This is true for both the Nikon’s all-phase AF system and the Sony’s hybrid system. Once continuous AF is confused, it must focus very quickly once the swimmer comes out of the water.

The Nikon setup I used is faster than the Sony setup. I don’t know how much of that is the camera, and how much is the lens. I have no way of testing, but I do know that autofocus speed is highly dependent on the lens. The older Nikon 80-400 was snail-slow, and the Nikon 400/2.8 is shockingly fast, both used on a D3 or D4. The new Nikon 80-400 is a lot better than the old one, but by no means in the territory of the 400/2.8. Still, it was fast enough to get the job done, and the Sony 70-200/4 wasn’t.

Autofocus speed may relate to  battery capacity as well. The motors in lenses could benefit from a lot of current for a short period of time to increase focusing speed, and that current may not be available from cameras with small batteries, like the a7RII. That relationship could be fixed with the addition of a supercapacitor, but I don’t think Sony would want to incur the cost, size, and weight penalty for what has got to be a corner case.

I did have some success with the a7RII by the pool. Breaststroke turned out not to be quite as challenging.

_DSC4648

_DSC4695-2

_DSC4700

 

There is another difficulty connected with using the a7RII on this kind of subject. I never could use single shot mode to pick the right moment to make the exposure. I think that is because of the latency introduced by the electronic viewfinder system in the a7RII. It seems to be even worse when the camera is operated in continuous low mode. If I’m right about the source of the problem, and it is not a subtle problem, the only solution for mirrorless cameras in this situation would be a faster viewfinder refresh rate.

 

Sony a7RII bulb exposure with LENR

There has been a little speculation that the increase in noise observed in the ISO 100 long exposure noise reduction (LENR) images in yesterday’s post might be wholly or in part the result of the a7RII’s dropping into 12-bit precision when LENR is active.

Fortunately, there’s a way to test that. Yesterday’s exposures were 30 seconds long, made in single shot mode with the camera timing the exposure. That results in 13-bit non-LENR and 12-bit LENR images. In bulb mode, the camera also drops into 12-bit mode (and performs some spatial filtering to boot.

So, if we compare two 30 second dark-field bulb exposures, one with LENR and one without, we’ll have two 12-bit images to compare, and we can see if the LENR still hurts the standard deviation.

A 400×400 central ample without LENR:

_DSC2077-Sel-3800-2450-400x400 lin

And with LENR:

_DSC2039-Sel-3800-2450-400x400 bulb lin

You can see the LENR exposure his higher “sidebands”, indicating more noise, although in both cases, the noise is close to the least significant bit of the 12-bit representation, indicating possilble insufficient dither.

Looking at log representations shows LENR at work. First, the non-LENR image:

_DSC2077-Sel-3800-2450-400x400 log

There is indication of a few hot pixels in the tail to the right.

Now with LENR:

_DSC2039-Sel-3800-2450-400x400 bulb log

The tails are gone, but at the expense of increased spread to the histogram.

Now let’s look at a few histograms of the entire image. The danger of doing this is that we could be looking at pattern read noise, but you could argue that LENR is supposed to reduce that, too. The advantage is that, instead of a measly 160,000 pixels in the sample, there will be more than 42 million.

Linear histogram, no LENR:

_DSC2077-Full-8000x5320 bulb

And with LENR:

_DSC2039-Full-8000x5320 bulb lin

The same spreading occurs as in the smaller sample.

Now with log scaling of the vertical axis. First with no LENR:

_DSC2077-Full-8000x5320 bulb log

With the larger population, the hot pixel tail is more apparent.

And with LENR turned on.

_DSC2039-Sel-3800-2450-400x400 bulb log

The tail is gone, but at the expense of a larger standard deviation.

Let’s put this noise in perspective. It’s so small encoded with 12-bit precision that LENR is unnecessary, even if it had no negative side effects. On the other hand, the damage done by LENR on these images in minimal, and probably inconsequential, once you’ve accepted the drop back to 12-bit precision that comes with bulb mode.

But with long exposures, LENR is a pain to deal with in the field. Best to leave if off at low ISOs. Even at high ISOs you should think twice about using it, especially since Sony starts using it a shutter speeds that are too short.

Sony a7RII LENR at ISO 100

Yesterday we saw that the Sony a7RII long exposure noise reduction (LENR, aka dark-frame subtraction) is applied at shutter speeds so short that it hurts the engineering dynamic range of the image. That experiment was performed at ISO 3200.

It just didn’t seem right that the Sony engineers would do something like that. To be fair, the right time to introduce LENR is a function of sensor temperature and ISO setting. I have only indirect control over the sensor temperature, so I thought to do a set of dark field images at ISO 100.

First, with no LENR:

no LENR ISO 100

Here’s how to interpret the graph. The horizontal axis is exposure time in seconds, with 1/125 second at the leftmost point, and 30 seconds (more or less) as the rightmost part of the graph. It’s a log scale, with each vertical line one stop apart. The vertical axis is the standard deviation of a 400×400 pixel central sample. This corresponds to the root-mean-square value of the noise, which is the most widely accepted, but not the only measure of noise. This is a fairly easy measurement to make in RawDigger if the camera does not subtract the black level before writing the raw file, which is the case with the a7x cameras. You need to tell RawDigger not to subtract off the black level, though.

The vertical axis is also a log scale, and shows how many stops the noise level is from saturation of the sensor/ADC system. Since less noise is better, you want the noise level as many stops below full scale as possible. Thus, values of the lines towards the top of the graph are better.

It is apparent that, at ISO 100, the a7RII doesn’t really need LENR. Let’s see what happens when you turn it on.

LENR ISO 100

The LENR noise levels are in all cases with shutter speeds of one second and longer worse than in the non-LENR case. Nothing but bad can come from selecting the  LENR at ISO 100 in the Sony a7RII.

Some caveats are in order. This applies at an ambient temperature of about 25 degrees C, with little self heating. It applies to one sample. And it applies if the noise metric is the standard deviation of the noise. Noise metrics that weight outliers more heavily might favor LENR.

A reader, in a comment to this post, raised the issue that the area that I selected for the graphing might not have any hot pixels, and therefor not get the benefit of the issue that LENR was created to fix, and suffer the damage of a half stop more uncorrelated noise from the dark frame subtraction, plus the coarser precision that Sony uses when LENR is used.

Here’s a histogram of the central region of the non-LENR image with a 30 second exposure:

_DSC2076-Sel-3800-2450-400x400 30 sec

The histogram looks pretty Gaussian, indicating that hot pixels may not be much of a problem.

With LENR:

_DSC2038-Sel-3800-2450-400x400 30 sec linear

Again, we just see the broadening of the histogram due to the increased uncorrelated noise and don’t see an improvement in the hot pixels that we couldn’t find in the original image.

But let’s dig a little deeper and look at a pair of histograms with a log vertical axis.

First with no LENR:

_DSC2076-Sel-3800-2450-400x400 30 sec log

Now we can see the hot pixels in the form of the extended rightward tail of the histogram.

When we invoke LENR:

_DSC2038-Sel-3800-2450-400x400 30 sec log

The hot pixels are cleaned up, but the histogram has a poorer standard deviation. LENR is doing what it is supposed to do, but the gain in hot pixel suppression is outweighed by the loss in increased uncorrelated noise and the drop in precision from 13 to 12 bits.

Does Sony a7RII LENR invoke spatial effects?

We saw in the previous post that the long exposure noise reduction (LENR) option in the Sony a7RII reduced the precision from 13 to 12 bits. We have previously seen that some 12-bit modes bring spatial filtering to the a7RII. Is LENR one of them?

Here’s the spectrum of a 1000×1000 sample from the center of a a7RII dark field image at ISO 3200 at a shutter speed of 0,8 second, with LENR on, but not used by the camera because it’s too fast.

p8 lenr not used fft

The vertical axis is the noise level in decibels with an arbitrary reference. The horizontal axis is frequency, with half the sampling frequency on the far right. A pure white noise spectrum would yield a ruler-flat straight line with an infinite number of samples, and a jagged one with fewer. Except for a very low frequency value for the horizontal component which is due to column noise, this is a textbook example.

Now let’s look at what happens with a 1 second exposure, which causes the LENR operation to take place:

p8 1 sec lenr fft

Not much difference. Let’s look at a 30 second exposure timed by the camera:

p8 30 sec lenr fft

Pretty much the same, too.

Now, a 30 second exposure in bulb mode, which we have seen to previously cause low pass filtering with LENR off. Are things different with LENR on?

p8 30 sec bulb lenr fft

No, they are not. This bulb exposure causes the high frequencies to be attenuated by about a decibel and a half compared to the low ones.

Sony a7RII long exposure noise reduction

I’ve had a couple of questions about the long exposure noise reduction (LENR, aka dark-frame subtraction) in the a7RII. I’ve had to do some research, since I never have used this function in any alpha 7 camera. It’s not that LENR is useless, just that its usefulness is limited, and that the kind of photography that I do doesn’t fit what LENR is good at.

What is LENR? The camera makes a picture when you snap the shutter. Immediately afterwards, it goes through the same sensor reset/wait/sensor readout operation that it performed for the exposure, but with the shutter closed. The second image is called the dark frame. Then the camera subtracts the dark frame from the real image and stored the result in the raw file.

What’s the point of all this? For long exposures — and what’s a long exposure in this context depends on the camera model and the temperature of the sensor — the dark-field noise is dominated by something that sensor engineers call “dark current” leakage in the back-biased photodiodes in the sensor that leads to an accumulation of electrons that are not due to light. The leakage current of the photodiodes is not constant across the sensor, and tends to be more-or-less the same for each pixel; in other words, it’s correlated from frame to frame.

Thus, subtracting a dark frame from the real frame can come close to eliminating the dark current noise that’s the same from frame to frame. It works best to subtract the mosaiced images, which is the way the camera sees them, so it’s not hard to do in-camera.

Sounds great, huh? Not so fast.

  • Done in camera the way I’ve described, LENR increases uncorrelated noise by the square root of two, or 1.414.
  • The time between successive exposures is doubled.
  • If the dark current for any pixel is so high that it saturates, dark frame subtraction will yield a value of zero for that pixel. A smart dark-frame subtraction algorithm could detect those pixels and do something else besides subtraction for them, but I don’t know what’s actually done in cameras.

For exposures that are into the dark current dominated region,  but not really long, I prefer to clean up the images in post.

I don’t do many exposures that really need LENR to look clean enough, but if I did, I’d do the dark frame subtraction in post, forming the dark frame to be subtracted by averaging 16 or 32 dark exposures, and thus eliminating the 1.414 times penalty for uncorrelated noise. I have used astronomical programs like ImagesPlus for this kind of thing in the past, and that’s what I recommend if you want to try it at home. I just write little Matlab scripts when I need to do similar things now.

How well does LENR work? Pretty well at really long exposures, but you actually take a step backwards at one second. Here’s a series of dark field exposures from 1/125 to 25 seconds with LENR on. By the way, LENR only works in single shot mode.

a7rII sh series lenr

Here’s how to interpret the graph. The horizontal axis is exposure time in seconds, with 1/125 second at the leftmost point, and 30 seconds (more or less) as the rightmost part of the graph. It’s a log scale, with each vertical line one stop apart. The vertical axis is the standard deviation of a 400×400 pixel central sample. This corresponds to the root-mean-square value of the noise, which is the most widely accepted, but not the only measure of noise. This is a fairly easy measurement to make in RawDigger if the camera does not subtract the black level before writing the raw file, which is the case with the a7x cameras. You need to tell RawDigger not to subtract off the black level, though.

The vertical axis is also a log scale, and shows how many stops the noise level is from saturation of the sensor/ADC system. Since less noise is better, you want the noise level as many stops below full scale as possible. Thus, values of the lines towards the top of the graph are better.

You can see an immediate half-stop increase in noise when LENR kicks in at 1 second.

Let’s compare it to a similar series without LENR:

a7rII sh series no lenr

By the 25 or 30 seconds, LENR has reduced the noise by about a stop over what you’d get if you didn’t use LENR.

Does LENR affect bit depth? Yes, indeedy. Here’s the histogram of a one second exposure:

_DSC3218-1sec

Only every fourth bucket is populated, indicating 12 bits of precision.

What if you turn LENR on and use a shutter speed faster than a second, so the camera doesn’t actually do dark frame subtraction?

Here’s a 0.8 second histogram:

_DSC3218-halfsec

Now every other bucket is occupado, indicating 13-bit precision.

Now let’s look at a 30 second exposure with LENR.

_DSC3232-Full-30

And one without:

_DSC0685-no lenr

This is weird. Look at the bell shaped curve of the LENR exposure, and compare non-LENR exposure. The spread is substantially larger for the exposure that has LENR applied. But, if you look at the graphs above, you’ll see that the LENR exposure has a lower standard deviation than thenon-LENR  exposure. How can that be?

The answer, I believe, lies in the outliers, the “hot pixels” which don’t show up on either histogram. You can get an idea of what they are by looking at the max values in both histograms, although that’s an imprecise and uncertain measure. There are pixels that are at fullscale in all four channels in the non-LENR exposure, and the maxima are substantially below that in the LENR exposure.

Remember what I said about the beginning about dark frame subtraction reducing the frame to frame correlated noise but increasing the uncorrelated noise by 1.414? I think that why we’re seeing the histogram spread even as the LENR is switched in.

LENR: use with caution.

 

Photography meets digital computer technology. Photography wins — most of the time.

Entries RSS
Comments RSS