DSLR Resolution and Noise Comparison
© 2006 KenRockwell.com. Also in Russian.
Resolution and noise have almost nothing to do with real image quality, unless you look at three-foot (1m) wide images from a few inches away. The details of this test only show the finest details, not the much more important color and overall look of these images. The real differences between cameras are in ease-of-use, color, and exposure. Wait until I show these real differences in another upcoming article!
See Why Your Camera Doesn't Matter for more.
As expected among highly competitive products, these cameras have very similar image quality. The more important differences among them are how they handle and how easy they make it to get this quality under real-world operating conditions.
I spent a week on this because I was curious enough to want to confirm or refute the old wives' tales.
Want to see how silly this all is? Roll your mouse over this image to see the difference between ISO 3,200 and ISO 100 with a D80.
Guide image, test targets behind stovetop. Roll mouse over to compare ISO 3,200 with ISO 100.
How much difference do you see? If you think this is broken, notice the clock's second hand and the lettering at the bottom left of the photos. I see very little noise at ISO 3,200. The other DSLRs did just as well. This is the full image, while the rest of the test examples are tiny crops of the little targets in the middle.
Because microscopic details are so easy to compare, far too many newbie photographers waste far too much time worrying about resolution and noise. Don't worry - none of the interesting things I'm going to show below matter much in real prints. Your screen shows even more than a print does at the same large size, because your screen is showing the actual image and not losing sharpness as most printing systems do. Your LCD probably has 100% MTF and a print doesn't.
If you enlarged any of the crops below as complete images at the same magnification, they would be over 38" (1m) wide!
I borrowed a Canon 30D and 20D and used my own Nikon D80, D200 and D70. Since the webpage layout has a spot for a sixth camera, I threw in my Canon SD700 pocket camera as the pig at this party to show the huge difference between compact cameras and DSLRs.
Light and Exposure
It was dim.
At ISO 100 the exposure was f/4 at one second. The Canons were shot at f/5.6 and a half second.
At ISO 3200 the exposure was f/4 at 1/30, or f/5.6 at 1/15.
I started this as a tough high ISO test in dim home lighting. The targets were lit with the dim orange bulb in the range hood, which matches dim interior lighting and requires a tough custom WB setting to compensate. This setting boosts the blue, which is most often the noisiest color. Luckily I see no problems in blue. This was at about 2,300 K, below the range of any of the camera's K settings, and far below the 3,200 K indoor presets.
Modern (2004 and newer) digital cameras smear noise at high ISOs with firmware employing varying degrees of cleverness. This is called Noise Reduction, or NR. NR smears the image to hide noise. Clever NR leaves the subject's edges and details alone to retain sharpness. Unfortunately NR also removes texture and details from the image as it smoothes over noise, even if it leave hard edges alone. The noise is always mixed with the image, so the NR has to be very, very clever to figure out which is which. Pushed too hard it makes the image look like a cartoon.
Because modern NR actively changes what it does from region to region in the image, digital noise performance no longer can be evaluated with a flat field as we could do with film. One needs to see what happens to the active image, not to a flat field, with digital cameras. Film grain is always the same everywhere in the image, but modern noise reduction acts differently depending on the detail in the image at each point. Read more at Fallacies of Noise Measurement.
Any idiot can design a camera with noise reduction that smoothes over everything and looks great on flat fields. That doesn't count, so I added a detailed radial resolution target to see if resolution drops at high ISOs. It's far more important to see what details and textures are left after the NR does its work.
I also grabbed a photo of my wife and I and dropped it in on the right. This is the best test to see what gets lost at high ISOs.
I shot this in front of granite to provide real texture that needs to remain in the image. Good luck to the NR firmware trying to tell the difference between noise and granite!
Noise Reduction Settings
My D80 and D200 allow selecting different levels of NR. The other cameras have no adjustments.
I left them at NORMAL so I could complete this report in less than a week.
I could have set them lower, which would leave in more noise and detail, or stronger, which would have smeared over more details and noise.
I have another page showing how these adjustments work at D200 NR Settings.
Focal Length (Skip if you don't care - it doesn't matter)
All the Nikons were shot with my fixed 28mm f/2.8 AI-s manual focus lens so I didn't go crazy trying to match zoom and focus settings.
I set the other cameras to give the same field of view. The 20D and 30D have slightly smaller sensors and their zooms said 27mm after I adjusted them.
My Canon SD700 was shot at 8.6mm.
Reproduction Ratio and Resolution (Skip if you don't care - it doesn't matter)
I shot at 5.6m (18.4'), or a magnification of 1:200 for the DSLRs and 650:1 for the SD700.
My star target's innermost ring is 1 Al/mm and the outermost ring is 0.2 l/mm (lines per millimeter).
For the DSLRs the outermost ring is 40 l/mm on-sensor (200 x 0.2). Note that 40 l/mm is usually the very highest resolution plotted on MTF curves. The next ring is at 60 l/mm the third is 80 l/mm. For the DSLRs the innermost ring is 200 l/mm on-sensor.
I see the Canon DSLRs hitting about 52 l/mm max before they start to alias back down diagonally (explanation later). The Canon DSLRs have about 60 l/mm resolution directly up-down or left-right.
I see my D80 and D200 matching each other at 60 l/mm overall, and about 70 l/mm up-down and left-right.
The SD700 has a much smaller sensor, so needs much higher on-sensor resolution to make the same photo. For its smaller lens and sensor, the outer ring is 130 l/mm (650 x 0.2) and the most the SD700 can muster diagonally before aliasing. Up-down and left-right I see 182 l/mm. Now do you see why compact cameras don't stop down to more than f/8? They need this insane resolution to make images with their tiny sensors. Diffraction limits resolution, and even at f/5.6 with the DSLRs I could see some slight effects.
More Test Procedure Details
Test results are meaningless if the method and setup aren't also published. In science and engineering we always publish that along with the results.
I cover more details at How I did this. Most people could care less, but the people who are kept awake late at night about these issues will care.
Next: NATURAL IMAGE NOISE TESTS (click for a whole page of examples)
RESOLUTION and NOISE TESTS (click for a whole page of examples)
I used my 28mm f/2.8 AI-s manual focus lens on my Nikons because using a zoom would have driven me even crazier attempting to keep the zoom setting and focus the same among cameras.
I used a 17-85mm IS len on the Canons. It had more than enough resolution to excite aliases in both the 20D and 30D.
I used the built-in lens with my Canon SD700 point-and-shoot.
I focused each camera manually to get the best focus, looking at the playback LCD.
My D80 and D200 offered perfect image detail and fidelity at 100% when zoomed.
Oddly, I discovered that the Canon 20D and 30D pull some funny business in playback and add resampling artifacts and lower resolution on the LCD. Look at one of these star targets on the Canon 20D and 30D at large magnifications and you'll see a different image than you'll see here or on your computer, and with a little less resolution.
My D70 didn't have enough playback magnification to check in-camera.
My SD700 played and enlarged its own files perfectly, but added the same additional artifacts to images played back from the 20D, 30D and added them to images from my D80, D200 and D70.
My 28mm f/2.8 AI-s manual focus lens is so sharp that I could see degradation from diffraction at f/8 and almost at f/5.6. I shot at f/4 for optimum results. The Canon 17-85 IS didn't open to f/4 at 27mm, so I shot it at f/5.6. As you can see, it had more than enough sharpness to excite aliases above the resolution of the 20D and 30D.
My SD700 used f/3.2.
Exposure and Manual WB
Each camera was set to the same manual exposure. I set a custom white balance on the gray card. Luckily for me, each camera was able to set to the very orange lighting (low Kelvin temperature) perfectly.
My SD700 pocket camera had no manual exposure settings, so I ran it in Auto. It does have great manual custom WB, and set it as I did the others.
I enlarged lower resolution images to match the higher resolution cameras: 3,872 pixels horizontally. Luckily Photoshop added and subtracted nothing in the process - it just made them bigger for a better comparison. This way we can compare all these cameras at the same equivalent print size: 38" (1m) across. (Photoshop has an easy time upsampling Bayer-interpolated images.)
The SD700's zoom, like every pocket camera's zoom, only works in set steps. I resized its images to 3,250 wide to give the best match.
Each of the crops was saved in Adobe ImageReady at 80%. Even at 40% things look almost identical. At 51 - 65% I can't see any real difference between the saved image and original, and at 80% there is no difference at all. You're seeing what I saw, but sorry for the long download!
D200: 0 Sharpening, 0 contrast, Color Mode III, Saturation +. Large JPG NORM, Optimal Quality.
D80: 0 Sharpening, 0 contrast, Color Mode III, Saturation +. Large JPG NORM.
D70: Optimize Image: Normal (whoops, this cancelled the saturation bumps I usually use, making it look duller), Large JPG NORM.
30D: 3 Sharpness, 0 Contrast, +4 Saturation, 0 Color Tone (all default except saturation).
20D: 0 Contrast, 0 Sharpness, +2 Saturation, 0 Color Tone (all default except saturation).
SD700: Vivid, Large JPG Super Fine.
SUMMARY back to top
All the DSLRs are the same overall. Each brand performed the same.
There are some minute differences if you look really hard at these extreme blow-ups. In real photography they will look the same as each other. If you have the time to waste worrying about these minutiae, you're probably very new to photography. Don't worry and go take some photos. The real differences between cameras are in ease-of-use, color, and exposure. Wait until I show these real differences in another upcoming article!
The ultimate noise levels are the same.
The DSLRs look great to ISO 1,600 and get a little worse at ISO 3,200.
Use any of the ISOs up to 3,200 with the DSLRs if you need them to get sharp pictures. The additional noise is minor compared to the increase in speed. Digital SLRs are better than film has ever been.
There is a huge difference between DSLRs and compact digital cameras.
I ran this test because newer photographers have been asking me for years worrying about the noise performance of Canon vs. Nikon and CMOS vs. CCD sensors.
Excuse me, but do you see anything significantly different among the DSLRs? The only thing I see is more aliasing with the Canons, but the same noise. I have a sneaking suspicion that the people who start these old-wives' tales start them from what they read in a press release from a camera maker, or an article written by someone who finds it easier to repeat the press release instead of running his own tests. These shots took me a couple of hours to design and a couple of hours to shoot, paying attention to maintaining the same focus and conditions and settings for all cameras, then a week to format into this web page.
As explained at Fallacies of Noise Measurements, any numeric values you're read are usually meaningless. You have to use your own eyes to evaluate noise in digital still images.
Any simpler effort introduces other variables which skew the results. I believe it is these other variables which lead to the old wives' tales of Canon or Nikon having better high ISO performance. Once someone expects a result, they (and I) will tend to see that result. It's simple psychology.
I see the same noise and the same real resolution from each modern DSLR, give or take an invisible 20%. The old D70 at 6 MP has a little less resolution as expected. Above the real resolution, I see much cleaner response from the Nikon D80 and D200 (just gray in the middle of the star target) and aliasing from the Canons. Since real photos almost never have any details as fine as this star target I call it a dead heat, of course with the SD700 pocket camera not included.
If you want to split hairs, the Canons have lower resolution than the Nikon D80 and D200, and that's how they're rated (8 vs. 10 MP). The Nikon D70 has the lowest resolution of these DSLRs, and is rated only 6MP. As you can see MP makes a difference, but not that much. That's the MegaPixel Myth.
If you find this as helpful as a book you might have had to buy or a workshop you may have had to take, feel free to help me write more with a donation.
Thanks for reading!