Archive for October, 2007

Noise-Density Illustrated

Sunday, October 28th, 2007

In our last post, we discussed the difference between noise and noise-density. Basically, noise-density is noise relative to resolution. Output quality, on a particular medium, is more related to noise-density than to noise because most mediums have an optimal resolution. For example, a typical computer display running at 1600×1200, needs about 2 megapixels. More pixels are simply wasted. The quality of an image seen at 1600×1200 is therefore related to the amount of noise that appears at 2 megapixels.

To illustrate the difference, I shot the same scene at 3, 6 and 12 megapixels using the Fuji Finepix F50SE. The experiment was repeated for various ISO sensitivities. The difference is striking: As the resolution decreases, noise decreases in proportion. What is a very noisy ISO 1600 image at 12 megapixels, becomes much less noisy at 3 megapixels. At ISO 400, the crop is noise at 12 megapixels but rather clean at 3 megapixels.

ISO 400
ISO 400 - 12 MP - Fuji F50SE ISO 400 - 6 MP - Fuji F50SE
ISO 400 - 3 MP - Fuji F50SE

All these images come from the same camera yet they show very different noise characteristics. The only difference is resolution. The top left image was shot at 12 megapixels, the top-right at 6 megapixels and the left image at 3 megapixels.

For an ultra-compact digital camera with a tiny sensor, the performance at 6 megapixels is excellent. At 12 megapixels, it does not seem so great, although it would take a relatively large print to take advantage of all those pixels. When using a medium that requires a resolution lower than that of the image, pixels are down-sampled to the optimal resolution anyways.

ISO 1600
ISO 1600 - 12 MP - Fuji F50SE ISO 1600 - 6 MP - Fuji F50SE
ISO 1600 - 3 MP - Fuji F50SE

Here is the same setup except at ISO 1600 rather than 400. While the 12 megapixels crop shows strong image noise, the 6 megapixels crop is good for an ultra-compact. Indeed, other than previous F-series Finepix cameras, no ultra-compact digital camera can produce 6 megapixels images that are this clean. At 3 megapixels, which is sufficient for 8″x6″ prints, only a little noise remains visible.

Note that these tests were made with the camera set to the desired resolution but nothing prevents users from taking their images at full-resolution in the camera and reducing the resolution later. Actually, this method is generally preferable because you can effectively reduce the resolution in software but not vice-versa. There are advantages though to reducing the resolution of images in-camera: reduced memory requirement and increased speed of operation.

Neocamera Blog Neocamera.com © Cybernium.

The Noise of Things to Come

Tuesday, October 23rd, 2007

Fuji Finepix F50 SampleMegapixels are increasing. Ultra-compacts such as the just-reviewed Fuji Finepix F50 now sport 12 megapixels sensors. Some people love it, some don’t. The controversy is mostly about image noise. More megapixels generally means more image noise, right? Well, it depends how you see it.

What we have to think about is our medium. How do we look at – not analyze – our pictures? How do we share them? For digital pictures, there are 3 common answers:

  • Upload them to an online gallery.
  • View them full-screen in a slide-show or screen saver.
  • Print them. 4″x6″ is still very popular. Up to 9″x12″ is common now.

If we think about these different mediums and ask ourselves how many megapixels are needed, the answer is less than what most modern cameras offer. For on-screen viewing, the most megapixels of any screen still in-production is 4 megapixels. And that would be for a 30″ LCD such as the HP LP3065. The typical monitor has less than 2 megapixels and so does 1080p HDTV. Prints are more demanding, but even a 300dpi 12″x9″ print only uses 9 megapixels.

So if we have a camera with too many megapixels, what happens to the megapixels we do not need? They do not simply get discarded, they get filtered. The difference? Discarding pixels is skipping pixels and making an image with the rest. Filtering is producing new pixels from a greater number of original pixels.

By filtering, we are affecting the noise-characteristics of an image. When an image is filtered to produce an image with less megapixels, noise decreases. The reasons can be explained by signal processing, but the details are not important for this discussion.

Consequently, the amount of noise a camera produces is not the same as the amount of noise appearing in our prints or on our screens. These noise levels are related, but they are different. Obviously, we want to judge a camera’s image quality based on our medium since that is our ultimate use for images. To this effect, we should not measure noise itself but something like noise-per-megapixel. Lets call this noise-density.


Noise is what we see when view an part of an image on our monitors because we called the image to 100%. Noise-density is what we see when view a photograph in our medium of choice. For large prints, noise and noise-density can be the same, but, for most prints and on-screen viewing, the former higher than the latter. Between a 12 megapixels image and a 4″x6″ print, the difference can be quite striking.

The conclusion from all this is that a camera should be judged on the quality of its output in the medium we use, not by 100% viewing. Viewing at 100% is informative and interesting, but alone it cannot determine wether one camera’s output is better than another. To judge which of two cameras produces better output, they have to be compared using the same medium. That means the same print sizes and the same screen resolution.

For the Fuji Finepix F50’s review, there was a hesitation on wether to give it a Good or an Excellent rating because its image noise is higher than that of its predecessors. However, after seeing that the quality of its prints were so good, the decision became clear: the F50 could not be punished for having that many megapixels if it ended up producing prints that were comparable in quality to other cameras rated Excellent.

Neocamera Blog Neocamera.com © Cybernium.

Olympus E-3 – Another 100% Coverage Viewfinder

Tuesday, October 16th, 2007

Olympus E-3With its introduction of the E3 today, Olympus is continuing a trend that I really like: more affordable DSLR cameras with a 100% coverage viewfinder. In fact, the E-3 is now officially the cheapest DSLR with a 100% coverage viewfinder, beating the Nikon D300 by $101 USD.

At the same time they increased the size of their viewfinders by providing one with 1.15X magnification. Remember that magnification is relative to the sensor size and that, although this is 15% more than previous Olympus DSLR models, this new viewfinder is not enormous compared to DSLR cameras with larger sensors. See Neocamera’s DSLR Viewfinder Sizes feature article for details. Look at the Effective Size column to compare size in absolute terms.

Neocamera Blog Neocamera.com © Cybernium.