Monday, October 29, 2007

G9: Noise in the Garden

Any self-respecting Pixel Peeper will take one look at this 100% crop and cry out “Noise!”. (You’ll probably have to click on the image to see a larger version and get the full effect.) On learning that the image is from the Canon G9, the Peeker will further exclaim, “I knew it. Small pixels are noisy. The G9 has too many pixels on a small sensor and therefore each pixel must be very small. No wonder the image is noisy; I predicted it.”

All of the above (explanation, that is) seems to be true. The Canon G9 does pack a lot (12,000,000) of pixels onto a small sensor. Therefore, those pixels are small and, all other things being equal, smaller pixels produce more noise. I’m not a camera designer, a chip designer or even an electronics engineer but I’ve come to accept these facts as being in accord with the “Laws of Nature” and, even more convincingly, Internet Lore.

Besides, the detailed tests and reviews of the G9 are published and it produces noisy images. No doubt about it: G9 images are noisy and the noise gets worse as the ISO setting increases. The full scene from which the noisy crop was taken is shown below (but reduced in size to 1024x768, high quality jpg compression).

Two of the main reasons that I bought the G9 were that I wanted higher resolution and a usable higher ISO than I could get from my Canon G3. The G3 was limited to 4MP and ISO400; however, the ISO400 was not particularly usable. My hope was that the G9 would have good quality at ISO400.

I’ve seen the reviews and read the comments in the discussion forums: G9 = Noise. I took a few shots at high ISO and peeked at the pixels. Yes, even my own G9 is noisy. Now, could I find a way to work with or around that noise?

Of course, I had to generate some images and conduct my own tests. Here’s the set up and procedure:
- found a scene with both bright and dark areas
- G9 on tripod
- Set for RAW + jpg (hoping to get improvements in Photoshop)
- Set Av mode
- Set 2 sec shutter delay (wish the G9 had a remote)
- Pictures at ISO 80, 100, 200, 400, 800, 1600

The above pictures were taken at ISO1600 and are direct from the in-camera jpg. Surprised? The second lesson learned is that the high ISO images can be very usable so long as the image is small. What good is a small image? Well, uh, the Internet...

The first lesson learned (see, the paragraph above has the second …) is that testing cameras is not at all straightforward. For example, what is the correct exposure for all these varying ISO speeds? I decided to use Av mode but that decision also means that the test results vary with the G9’s judgement of exposure. (Shutter speeds selected by the Av mode did not exactly track ISO.) Is the in-camera jpg that is embedded in the RAW file the same as a jpg made without RAW? (Seems to be.) How should the data be presented?

Entire scenes are shown in this post as 1024x768 pixels. The in-camera jpg images are presented as-is and 100% with no post processing whatsoever except for cropping to emphasize the noise.

The ISO 80 crop from in-camera jpg

The ISO 800 crop from in-camera jpg

The ISO 1600 crop from in-camera jpg is the first picture show in this post.

I’ve already admitted to peeking at a few pixels myself. I didn’t care for the higher ISO images from the in-camera jpg. In fact, I didn’t particularly like the ISO 80 image from in-camera jpg – not that it was noisy, it just doesn’t look good to me. Time to break out Photoshop CS3 and process those RAW files.

Interestingly, Adobe Camera RAW (ACR) did not automatically change the exposure of the RAW files. Of course, I did tinker with the settings myself. I wanted to minimize noise and get a picture more to my liking as well.

In the G9 review on Digital Photography Review, I was impressed by the noise reduction of an ISO 400 image with ACR 4.3 (and by noting that ACR 4.3 is being Beta tested; the current version is 4.2 and it only approximately fits the G9). The ACR settings really caught my eye: luminance at 0 and chroma at maximum. I’d have never thought to use these extremes but it seemed to help the DPReview images. Of course, I don’t have the 4.3 Beta but I tried both extremes and the middle settings for luminance and chroma noise reduction in ACR 4.2. To my eye, there was no effect; perhaps this will change in version 4.3.

I use Photokit Sharpener from Pixel Genius for sharpening. PK Sharpener uses a three step process: 1) capture sharpening, 2) creative sharpening and 3) output sharpening. Capture sharpening is user selected according to camera resolution and user taste, creative sharpening is entirely user taste and output sharpening varies with file size, output device (screen, printer, etc.) and user taste.

The most recent versions of Photoshop include sharpening in ACR. The ACR sharpening parameters are on the Detail screen along with the noise reduction parameters. I’ve been wanting to try ACR sharpening as a sort of “capture” sharpening. Strangely enough, to me anyway, when ACR sharpening is used, the luminance and chroma settings come into play. I suppose ACR is designed this way but, if so, it was not obvious to me. After much trial-and-error, it seemed to me that settings of 50, 0.5, 20, 80, 50, 50 for the ACR Details page reduced noise nicely but not entirely. (You’ll have to check the Details page to see what I mean by these numbers.)

I still used PK Sharpener for capture sharpening (incorporated into the trial and error above).

Now for more noise reduction using Neat Image. I made my own profiles using the Neat Image calibration targets and applied Neat Image at 35% strength. Pretty simple and took me a lot less time and experimenting.

By this point, the noise level had dropped quite a bit. I finished up the post processing by tweaking levels, curves and saturation. Here’s the 100% illustrative crop from the ISO 800 image:

Notice that the noise, even pixel peeked, is not too bad. On top of that, the details look better to me. And here’s the final, full image from the ISO 800 RAW file after post processing.

I even like this one more than the ISO 80 image from the in-camera jpg but that’s a matter of taste. I will say that the colors, especially the fence, are more accurate (if somewhat saturated). It made a nice print.

So, I think my G9 has very good potential at ISO 400 and even 800 under the right conditions and when processed from RAW. The requirement to process from RAW is not a problem to me; much of that processing can be done semi-automatically using Photoshop Actions.

But, sad to admit, I gave up on the ISO 1600 file. Although it could be improved, I haven’t got it quite right – yet.

No comments: