70

I've been experimenting with low light photography. I've photographed something weird inside my camera and don't really know what it is. This is the image initially taken...

original dark frame

Seems inoccuious enough. After some processing though, I found the following...

enhanced frame

What is it?

Some backgound. This was taken with a Nikon D80 set to ISO 3200 and a timed 30 sec exposure. Image format was RAW. There was a f2.8 - 4, 24 - 85mm zoom lens fitted (bulky - and I'll get to this later). The image was equalized and curve stretched by (a) GIMP. All images are /were RAW or PNG format so there shouldn't be any JPEG artifacts in there. So I have some questions.

  1. What causes the white bit at the top of the image? It is not light bleed. The image was taken with the lens cap on, with the camera inside a steel filing cabinet drawer, in my unlit garage at night. It is not light bleed. Further, to confirm this I actually took 2 frames, turning the camera upside down onto its' view finder. This white bit is always towards the view finder.

  2. What is the partial circle apparent in the bulk of the frame? Is this a thermal image? I believe that CMOS sensor based cameras can see a little into the near IR band. The lens is quite bulky and might have been warmer than the main camera body. Is the answer to question (1) also thermal. The top of the camera has a lot of controls and bits there.

  3. What causes the vertical lines in the image? Is this something to do with the Bayer filter or the linear addressing of the individual CMOS photodetectors?

  4. What causes the purple and green patches? I realise that these are not true colours, but nethertheless they are representative of macro scale intensity variations across the frame.

Any thoughts?

  • 2
    (Re adding my comment because it is not an answer, and if it was answered by the OP, I didn't see it before the comments were cleared.)I opened the top black image in Office 2010 Picture Manager and through a number of automatic and manual manipulations I saw something more similar to clabacchio's processed results, but colors more similar to the OP's results. Is there any chance the circular pattern is due to the post processing? – Mark Hurd Feb 15 '15 at 4:49
  • May not be relevant, but which aperture setting did you use? – clabacchio Feb 15 '15 at 15:16
  • Very late to the game, but I wonder, was there any lens correction applied somewhere in the RAW processing chain? – scottbb Dec 18 '18 at 18:13
36

First, understand a couple of things:

Even though we call these things "digital cameras," the process of turning photons into numbers is entirely analog. Analog circuits pick up all manner of noise from their surroundings.

Noise isn't one constant value, it's a range of them that top out at a level called the noise floor. The processing you did on the otherwise-black image stretches everything below the noise floor, most of which is close to black to begin with, across the entire dark-to-bright spectrum in each channel. The histograms before and after processing will show that.

What causes the white bit at the top of the image?

My guess is that it's thermal, perhaps the bottom part of the camera is a bit warmer for some reason than the top. Your processing will exaggerate this.

What is the partial circle apparent in the bulk of the frame?

That could be anything. The sensor could be picking up IR the low-pass filter missed and you're looking at your lens cap. It could be the result of deformities caused when the sensor was picked up from the back with a suction cup during assembly or minute amounts of stress placed on it by the retaining parts. I'm sure Nikon knows, but given how far down in the noise it is, it's either not worth caring about or it's been accounted for in the design of the rest of the camera.

What causes the vertical lines in the image?

Under these conditions, the best theory would be that some clock circuit in the body is inducing a bit of current onto one of the analog parts on a regular basis. If you check out this D80 IR conversion tutorial, you'll notice that there are components on the same board as the sensor and others close by on the main board behind it. There are lots of opportunities for that kind of leakage.

What causes the purple and green patches?

The sensors in digital cameras are noisiest in the blue channel, and if you look at the RGB composition of purples, there's lots more blue than red or green. The greener patches may be a result of some parts of the camera being noisier in different ways. Again, your processing exaggerates these differences. Most of the colors are actually so close to black that you'd never notice them otherwise.

  • 2
    I like the idea of the suction cup! – Fumidu Feb 12 '15 at 15:52
  • 3
    @Nick: But the blue channels is usually less sensitive, so has more amplification, so lower signal to noise ratio than the other channels. – Olin Lathrop Feb 12 '15 at 20:08
  • 1
    @Nick: It might work that way. Actually I wouldn't be surprised if there were 3 separate gain stages with 3 separate A/Ds. Switching the gain dynamically per pixel would be too slow. Not changing the gain could lead to low resolution in one or two of the colors. I haven't been envolved with this kind of sensor design, so I don't know how it is handled. It could possibly be handled differently between sensors. These rae all engineering tradeoffs available to the sensor designers. I don't know what is customery. – Olin Lathrop Feb 12 '15 at 22:52
  • 1
    +1 this is the correct answer. EM noise is all around us. The sensor+circuitry will necessarily pick up all sorts of noise, from both inside and outside the camera. Heck, the laws of physics require that any electric-based sensor generates noise which it itself will pick up. That's just how electricity works. – BlueRaja - Danny Pflughoeft Feb 12 '15 at 22:59
  • 3
    @ColeJohnson: Part 15's must-accept-interference requirement is for interference from other Part 15 devices. Last I checked, the FCC had no jurisdiction over Mother Nature and basic physics. :-) – Blrfl Feb 13 '15 at 11:45
23

Maybe it represents the small variation of the temperature of the sensor.

A hot sensor produces more noise than a cool one.

The small temperature difference can be explained by the presence of electronic components, or the way the sensor is in contact with other parts, allowing more or less heat dissipation.

Some related links :

http://en.wikipedia.org/wiki/Image_noise#Sensor_heat

http://www.zodiaclight.com/astrophotography/chipHeating5.htm

https://thephoblographer.wordpress.com/2010/01/22/freeze-your-camera-get-better-high-iso/

(I'll probably try this myself, see if we can obtain the same pattern!)

Edit

After some discussion in the comments, it appears that the sensor itself may produce some significant amount of heat (thanks to Davidmh and Chris H).

I have tried that with my camera. Olympus EM-1, Iso 200, 60s, F2.8. Raw image then processed in Lightroom. The camera was in total darkness of course, at room temperature (18°C) Here is the result: Olympus EM-1, Iso 200, 60s, F2.8.

I would say... well, not very conclusive! Can anybody see anything? Of course, it's a different camera. We should try with others D80 to see if it's specific to this model.

  • 1
    I have seen similar circular patterns on an astronomical CCD with active cooling before being completely cool. I would imagine the sensor at hand has a dissipator and it is probably best attached in the middle. It also reminds me of a CPU heat sink: ifisc.uib-csic.es/nuredduna/2004/fotos/cc_radiator.jpg – Davidmh Feb 12 '15 at 11:36
  • 3
    I see the pattern too. It reminds me of shear stress, that would be compatible with the glue hypothesis. Alternatively, it could be something like the thermal paste in a CPU heat sink (the grey bands). – Davidmh Feb 12 '15 at 14:03
  • 2
    As @Davidmh says this is certainly an issue with CCDs. when using webcams modified for astronomy (or other low-light imaging) the heat from the on-chip amp could cause a region to look illuminated. And I was surprised to see that according to dpreview the D80 uses a CCD and not a CMOS sesnor. That suggests self heating contributes. – Chris H Feb 12 '15 at 16:45
  • 2
    Off topic, but "Glue hypothesis" totally sounds "Big Bang Theory" to me... :-) – Fumidu Feb 12 '15 at 23:41
  • 1
    It's likely to be very different on a CMOS sensor such as your Olympus. CCDs are rather rare in SLRs so replicating on another camera may be a little tricky to arrange. Your image looks like pure noise to me - and I've spent quite a bit of time in the last few years trying to find faint spots in noisy images from both CCD and CMOS scientific cameras. – Chris H Feb 13 '15 at 12:40
11

This is basically noise, but from several different mechanisms. Consider the extreme amplification of small details you had to apply to get this picture. There are several distinct source of noise here.

The overall graininess is random noise from individual sensels. This is going to happen. Every sensor has some finite random noise added to whatever signal it is measuring.

The vertical streaks are sensor pattern noise. Nothing is ever exact, including the geometries and other parameters of the sensor as it was manufacturered. The sensor is a rectangular array, and some types of position and size errors of the matrix will show up as vertical and horizontal streaks. Why yours shows more vertical streaks probably has to do with the exact structure of that sensor. Note that there are some horizontal steaks too, but less evident.

The white part at top is probably a small thermal difference between the bottom and top of the sensor. Again, you have greatly amplified the tiny noise signal, so this can result from a small fraction of a degree C. The electronics have to be somewhere in the camera, and will cause a little more warmth there than elsewhere. Some thermal gradient across the sensor is impossible to avoid, and is not out of line at this extreme magnification of the noise.

The overall "double D" pattern is the inside of your lenscap. It's actually quite a testiment to the low signal to noise ratio of this sensor that it can image th inside of the lens cap at all. Take a look at the inside of the lens cap and you will see this pattern. Try the experiment again with the lenscap rotated 90° and you will see the large scale image rotated by 90°.

You may be wondering how it was able to image the lenscap when you took great care to eliminate light from around the camera. The answer is that it isn't visible light imagine the lenscap. The image is formed by near IR, and comes from small temperature differences and black body emission characteristics from different parts of the lens cap. Black body radiation is always around us, and is emitted by every object that isn't at absolute 0 temperature. This includes the internal structure of your camera, the lens, and the lens cap. Objects at normal human temperatures emit so little black body radiation, and the radiation they do emit is at such long wavelength, that we don't notice this. Normally your camera doesn't notice it either because the light from the scene is many many times stronger and overwhelms the received black body IR. In this case, you went to great lengths to eliminate that ordinary scene light and amplify the tiny signal that was left, and you can just barely make out the image.

Black body radiation is unnoticed on a human scale at human temperatures, but can be significant for some scientific measurments. This is why the sensor in high end telescopes are actively cooled. The Hubble space telescope uses liquid helium to keep the sensor at about 4°K if I remember right.

  • 2
    I'm sceptical that it's the lens cap imaged in IR (it's way out of focus and badly focussed by a visible lens anyway). But you've provided a way for the OP to test, so hopefully we'll see. – Chris H Feb 12 '15 at 16:47
  • @Chris: You may be right. I just tried to reproduce this with my Nikon D3s and failed. I used a 24-120 zoom at 24 mm, lens cap, 30 seconds, f/4, ISO 12800, eyepiece shutter closed, in a closed metal cabinet (achieved with 5s self-timer). I took two images with the lens cap rotate 90 deg. Both images just look like random noise as you would expect, no large-scale feature like in the OP's picture. – Olin Lathrop Feb 12 '15 at 20:05
8

Is the lens cap the center-pinch style? Seems like you can see the contours of the cap in green. If so, the other visible shapes (circle) are probably due to lens features as well. Try removing the lens and shoot with a body cap attached to see how the image changes.

The white portion could be light (IR, maybe) leaking in through the view finder. My camera strap has a rubber bit that fits over the view finder to prevent small light leaks during long exposures. Try covering the finder and see how that changes the image.

  • 11
    I am not sure you would get an image of that. The cap would be extremely out of focus. My money is on thermal homogeneities on the sensor. OP: can you please take another image without the lens? This should be sufficient to false one of the two hypothesis. – Davidmh Feb 12 '15 at 11:41
8

I recommend this paper for possible explanations. It’s about the Pancams used in both Mars Exploration Rovers. It describes a thorough examination of noise sources in CCDs and describes the detailed calibration routine for the Pancam CCDs.

Here is the Link to the paper: Mars Exploration Rover Athena Panoramic Camera (Pancam) investigation (PDF Document)

Another paper that might be interesting: In-flight calibration and performance of the Mars E xplorationRover Panoramic Camera (Pancam) instruments (PDF document)

Edit

After playing around with the original and processed images in ImageJ I tend to the conclusion that the radial structure is possibly an artifact of the processing itself because it appears mainly in the green channel of the processed image. There is a slight hint of a ring in the red channel, too. It would be interesting what your processing steps were.

Here is what I get if I contrast enhance each channel individually to its min and max value, vs. OP’s version:

Different Processing Results

I cannot see any similar structure in my version, except the vertical streaks and the lighter upper border of the image.

The fact that the ring pattern appears in the green channel but barely in the red channel speaks against the IR hypothesis because if I’m not mistaken IR would rather appear in the red channel. IR should at least appear with a similar intensity in the red channel, unless there is an IR leak in the green filter, but not in the red filter.

2nd Edit

I think the “original” version shown here lost the information during the conversion to PNG format. It’s impossible to recreate the weird pattern visible in the processed image. The closest I could come to the processed version is this:

Heavily clipped version of contrast adjusted image

First I maximized the contrast without clipping the image. Then I created a 3d surface plot of intensity values, clipped to the values of the enhanced image to a range between 66 (26% of 255) and 26 (10% of 255). There seems to appear a slightly circular pattern similar to OP’s processed version, but I guess the rest of the pattern is lost due to the lower bit resolution of the PNG file.

The original version, contrast enhanced, looks like this:

Contrast Enhanced Original

  • 2
    Please elaborate on what these papers would explain about the image. – Hugo Feb 13 '15 at 23:45
  • 1
    The papers explain in great detail how sources of noise (temperature dependency, read noise, full well capacity variations, digitization noise, dark current (thermal noise) and flat field irregularities) were analyzed and compensated in the calibration pipeline. Especially the dark current and flat field chapters came to my mind when I saw the image. The regular pattern seems to be caused by the photomask and surface structure of the CCD, for example. If I process the black photo then I get a result that has a character similar to clabaccio’s result. The radial pattern does not appear at all. – M L Feb 15 '15 at 3:14
7

Luckily I have a D80, so I tried to replicate your experiment. Unfortunately:

  • It was during the day;
  • I was in my garage, though it has a window from which some light leaked inside;
  • My cabinet is wooden, not steel;
  • My lens is a Sigma 18-50 f/2.8, unlike yours ;)

But I did use the lens cap, I did close the cabinet, and I set the aperture at f/22, just in case. I took two shots at 30s just in case, and got similar results for both, but different from yours:

This is the RAW as it came out of the camera (resized at 600px wide): enter image description here

Here I applied +4.5 EV exposure compensation on RawTherapee, to make the artifacts come out: enter image description here

It does show the white-ish glow on top, perhaps less pronunciated just because I didn't go all the way with contrast. It also shows the vertical bands, but I can't see the circular shape.

I agree with the theory about non-uniform heating on the sensor, and the bands could come from the CCD technology of the sensor. Depending on how it's build, the charge transport mechanism could cause the streaks.

  • To see if there is any way of imaging the inside of the lens cap, you would have to open the aperture all the way, not close it. It might also be interesting to cool the camera and/or warm the lens cap before the test. That would maximize the chance of imaging the lens cap in IR. – Olin Lathrop Feb 14 '15 at 15:05
  • @OlinLathrop well I didn't imply it was the image of the lens cap, my procedure was aimed to detect anomalies in the sensor. Would I try to examine the footprint of the lens cap, I would definitely need a darker place to prevent leaks. – clabacchio Feb 15 '15 at 15:14
-3

The image was taken with the lens cap on

What is the partial circle apparent in the bulk of the frame?

It's the lens cap.

  • 9
    While this may be correct, it would do better with the suggestions of how to test that theory (repeat with lens cap rotated) and possibly an attempt to reproduce your self. – user13451 Feb 12 '15 at 14:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.