What sky noise?

Some users of the Olympus Camedia E-10 camera raised the problem of what they consider excessive CCD noise present in daylight sky areas. (To be accurate, this is not limited to sky; sky shots just show the problem best. Really we are dealing here with the issue of CCD noise at reasonable, i.e., easily handholdable, exposure levels.)

What is "excessive" is a highly subjective concept. So far I'm finding the noise I can see on my E-10 acceptable for prints up to 11x14 inches (full frame).

This is a simple attempt to evaluate the magnitude of the problem, if any, and to investigate some other aspects of the noise on the E-10. Users of the Olympus C-3000Z and C-3030Z cameras may find this of some interests, as some data on the noise in their cameras is also presented here.

For a general introduction to the noise issues, refer to another article at this site.

Technical notes

Unless otherwise stated, all presented samples have been obtained under the following conditions:

• Recorded in the lossless TIFF mode; ISO speed set to 80 (E-10) or 100 (C-3000Z);
• No in-camera postprocessing, i.e., sharpness set to "Soft" and contrast (in E-10) to "Low";
• For Web presentation, uniform fragments of 160x160 pixel size were saved as JPEGs with zero compression, to avoid any artifacts.

Note that this size corresponds to a square of about 19 mm (0.76 in) size when a full frame from the E-10 is enlarged to an 8x10 inch format; for the C-30x0 the size would be 21.2 mm (0.83 in).

Samples taken from film have been transferred by Kodak to the Kodak Photo CD format (2048x3072 pixels). Our standard enlarged sample would be 15.9 mm to a side.

Enhanced samples: to make visual comparisons easier, I'm presenting, often in addition to the originals, samples with the brightness range of 32 (on the 0-255 scale common to image processing programs) stretched to the whole 0-255 interval. The 32-unit range to stretch was chosen so that non-zero bins of the histogram would start at, or immediately above, its lower limit.

For example, if a sample histogram had non-zero values between 120 and 135 (quite typical for a cloudless sky), the interval from 120 to 152 was stretched to 0-255.

How strong is the effect?

Indeed, the E-10 shows more noise than C-3000Z. These are two different CCDs, made by Olympus and Sony, respectively, so they may display different behavior.

The film noise (being a sum of film grain and the electric noise of the Kodak scanning device) is, however, even slightly greater; an effect which in printing will be offset, in part at least, by a higher scan resolution.

In-camera Processing

Many users set their cameras to the default "Normal" (as opposed to "Soft") settings of sharpness and, in case of the E-10, contrast. It is expected that this may result in greater amount of noise.

At this moment we could finish our discussion of the "blue sky noise", drawing two conclusions:

1. The E-10 displays more blue sky noise than the C-3000Z (and, probably, many other digital cameras);
2. This amount of noise is equally acceptable (or equally objectionable) as that in a 100 ISO negative processed on a high-quality, professional scanner.

Additionally, we can say that the E-10 noise increases, although not drastically, when the image is postprocessed in the camera as per "Normal" settings.

Those interested in further analysis of the noise phenomenon in the E-10 may stay with me and have a look at some further comparisons.

How much noise comes from the sky itself?

At first, this sounds like a crazy idea. Isn't noise a purely electronic phenomenon? We can not be sure, however, how much the fluctuations in light scattering may affect the uniformity of a blue sky. Probably not a lot, but it does not take much to check.

Do filters affect the noise?

According to some users' impressions, the E-10 "sky noise" seemed more visible when a UV filter was used. This would be quite difficult to explain. Here is a comparison of three "stretched" samples shot under identical conditions with no filter, with a UV filter, and with a circular polarizer.

Some of the E-10 users suspect that the noise may be mostly in the near infrared. This would be easy to check: there is a filter (B+W #486) sharply cutting off the UV and IR wavelengths just outside the visible range. Unfortunately, I don't own this filter, and I'm quite unwilling to shell out $102 (via B&H special order). Besides, the IR sensitivity of the E-10 is just marginal (see my article on infrared photography), so I'm remaining most skeptical on this possibility.

Adding all this together we may conclude that the noise seen in our images, sky or not, is of purely electronic nature, or mostly so. Its magnitude is comparable to that of an ISO 100 film. It may be more visible on a computer screen, when the image is viewed at a 1:1 pixel magnification. You will not see much, if any, of it in prints seen from the standard viewing distance close to the print diagonal.

Related pages

• My noise article with a general introduction to the problem
• My article on dark noise and hot pixels in the E-10
• List of all my pages related to the E-10