A little about DxO Mark

Since there has been a lot of interest in this camera after previous posts, I wanted to mention that DxO published their rating of the Nikon D3100 sensor.

DxOMark - Compare sensors

This chart compares the new Nikon D3100 to the highest-scoring APS-C sensor (Nikon D300s) and full-frame sensors (Nikon D3s).

I’ve used DxOMark a lot in passing discussions, and a friend (and recent Nikon D3100 owner) last month mentioned, “I don’t know how to read this.” Whoops!

I had better explain what these values mean.

The Overall Score is an aggregate based on the other scores normalized against some baseline value. It really serves no purpose other than bragging rights. The only important thing to know here is that this (and all DxO scores) are a measurement of the sensor and RAW files—so in-camera JPEG processing, and camera and lens performance are not a factor in any of these.

The Portrait score is the number of bits of color the sensor can distinguish at the base ISO under ideal conditions. This means, for example than the Nikon D3100 can distinguish 6 million colors (222.5). As a reference point, the human eye distinguishes somewhere under 10 million and your monitor renders 16.2 million.

The Landscape score is the number is f-stops of dynamic range the sensor can distinguish at base ISO. Dynamic range is the total tonal range from deepest near-blacks in shadow to the highest brights in highlights (without blowing them). This means, for example, that the D3100 can distinguish 11 f-stops while the Nikon D300s and Nikon D3s give you an extra f-stop. Remember, exposure bracketing can extend this range, and that not all ranges are equal (the shadows have less bits than the highlights, but the highlights clip) as mentioned in the article linked. So, this measures the absolute range, but the effective range is going to be less.

The Sports score is the “real” ISO of the camera, and here’s how you read it: the noise level of the ISO score are the same across cameras. This means that ISO 919 on the Nikon D3100, ISO 787 on the Nikon D300s, and ISO 3253 on the Nikon D3s have equivalent noise. In other words, if you never shoot higher than ISO 800 on the Nikon D3100, then you should be just as happy with the noise performance of ISO 3200 on the Nikon D3.

The last score shows very extreme differences between makes and models. And you may be disappointed with the Nikon D3100’s performance, but that’s because we’re comparing it to a $4000 body. Instead, let’s compare it to the high-rated “pro” level pocket digicam: the Canon G11. Hmm, that ISO 919 is an anemic ISO 161 on this camera—one which people rave about its low light performance. Standards for pocket cameras are different!

Now back to the Overall Score. While the ISO performance has gone up relative to the last of the previous generation designs, the Color Depth and Dynamic Range has gone down. How is that possible?

The reason is that megapixel has gone up while the sensor size hasn’t changed. What happens when megapixel goes up is the pixel pitch increases. This may mean a drop in ISO performance (and why big sensors like the D3 smoke the D3100 which smokes the G12), but there are various tricks Nikon is doing to get around it. What can’t be avoided, however, is that the electron well that collects the photo counts also gets smaller. With a smaller well, the accuracy of the counting goes down and so does the dynamic range and color-depth. Which depends on the engineering tradeoff in the sensor design.

This also explains why Nikon cameras, with their more modest megapixel, score higher than their Canon counterparts. The DxO score doesn’t reflect the megapixel and higher pixel counts are penalized with lower DxOMark scores.

Think of it this way. When purchasing a camera, people focus too much on the megapixel, the DxOMark complements that number with its own. The combination gives you an idea of what sort of sensor is in the camera you are purchasing.

6 thoughts on “A little about DxO Mark

  1. So what's the best way to compensate for the pixel count? Say for example that we scaled a photo from each camera down to 1 megapixel, then compared the files. Is there a way to adjust the dxomark numbers for this? Then you could ask questions like "if I scale a d3x 24megapixel image down to a 12 megapixel image, do I get the same low light performance as the d3s?" and "if I scale my picture by this much, how does it affect the sports score iso?"
    It's a relevant question, because a standard print will be made by scaling an image to a standard megapixel count.

    1. If you scaled each photo down to 1 megapixel then the megapixel resolution is irrelevant (see the "megapixel myth" linked in the article above) and the DxOMark is the only score you need for evaluating sensors.

      I don't think there is an easy answer for this. Sensor quality is not along a line but along different dimensions. Just think of these three DxOMark scores as three of the dimensions and pixel count as another dimension.

      As for print, it depends on the print size and viewing distance. If the viewing distance is 1 foot, for an 4×5" print, 2.5 megapixels is more than enough (and if your scaling is anything BUT photoshop, scaling down to it will look good too), 8×10" blow up needs 6 megapixel, a 10×16" double-truck needs at least 12. Anything more than those numbers is for cropping.

      1. Dynamic range wouldn't change — but NOISE would be reduced if you shrank a large MP file to match its lower MP counterpart camera's images (Canon vs. Nikon. Or D3x vs. D700). Therefore, the ISO score would improve for the higher megapixel camera compared to the DxO measure. So DxO is not perfect.

        But you can think of it this way — if the MP is a huge gap (say 12mp vs. 24mp), but the ISO is a tiny gap, then you should go for the more megapixels. If the Dynamic Range, however, has a huge gap then go for the one with the bigger dynamic range.

        1. @Global

          Hmm this is possibly a good point, but more likely not. It depends on the testing conditions used by DxO. Let me explain.

          DxO "Sports" score is a measure of ISO noise. The measure they use is signal-to-noise ratio. My thinking is they standardize the signal so it has the same "size" on sensor. For instance, if the signal is the letter "E" (like a giant letter E), they would make it so that letter E takes up the same percent area on the sensor they are testing. Thus, larger sensors would have lower noise.

          However more megapixels in the same sized sensor will not have more "noise" according to this measurement. Because what is being measured would be the amount of noise relative to the signal at the same spacial frequencies necessary to resolve the E, not at the spatial frequency of the sensor.

          This is a fair test do you not think? I’d be hard pressed to think comparing noise at the sensors inherent spatial frequency is fair. It also would be hard to test since very few lenses have any aperture with near perfect performance at the spatial frequency of the sensor (no lens, afaik) and thus your test would be highly lens performant. OTOH, it’s pretty easy to find an aperture where any decent lens will have resolution maximized and diffraction effects minimized.

          If this wasn't the case, I fail to see how large sensors with lots of megapixels continue to outperform smaller sensors with less (I'll admit it's hard to control for this since less megapixel sensors tend to be older).

          More megapixel decreases electron wells but increases the number of points to tease the signal. One thing makes things worse; the other makes things better. But, the better does not occur faster than the worse in many domains. In this case, if we are in a high ISO noise which is electrically gained near the limits, then we are talking about shot noise being dominant and innacuracies due to flaws in the electronic or optical systems are magnified by amplifcation.

          Because of this (very likely) possibility, I’d argue that my original view holds. If you are looking at an uncropped image and have a final format in mind, then if the sensor beats the megapixel that is "enough" then you can just use the DxOMark score head-to-head for sensors.

          However, images tend to get cropped. We don’t know the final format size of the image. There is a little something called a lens that is a factor also.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.