Sony Zone


THE DSC-F828 CAMERA is known to be "noisier" than others in the Advanced Cyber-shot line. That's true, but it can be understood--and helped!

Its pixels--actually the sensors on its image chip--are almost exactly the same size as those on the image chip of the DSC-V1, 5-megapixel camera. So comparisons are made here between those two.

Noises Off!

By Peter iNova  
Layout by Peter iNova

With every great stride forward, an equal and opposite stride often waits in the wings, ready to shove things backward. Think I'm kidding? Explain your alternate theory to the tax man.

With the DSC-F828, 8-megapixel supercam, comes a new imaging chip. It has 3264 x 2448 pixels in an array described as a 2/3-inch package. Of course, that doesn't mean its image sensor size, being a throwback to imager references that began with TV tubes in the 1950's, but it does mean that the chip in the 828 is the same physical size as the one in the 707/717 cameras that are similar in format and physical volume.

The image chip in the V1 is smaller, even though its pixel count is slightly larger than that of the chips inside the 707/717s. By shrinking the image sensor to about 83% of its prior technical manifestation, the surface area becomes only 69% of the original, and if the scaling process is extended into its depth (it may not be) that would imply a sensor that occupied just 57% of the original volume.

At the micro scales these sensors operate, that's a significant opportunity for the new smaller sensor to potentially not perform at the larger sensor's benchmark level. Any deviation of performance can show up as electronic noise--literally a failure of each sensor to perform at an ideal spec.

The sensors in the 828's image chip are about 79% of those in the 707/717s, so each has a surface about 62% of the former image chip's size. Very comparable to those in the V1. Of course the newer camera has the RGB+E color filter array (CFA) that may or may not play a part in any of the discussion that follows.

There are no perfect sensors. Not even the Hubble Space Telescope has zero noise, but on that digital camera, it chills its chip down to liquid Nitrogen level to help wrangle the noise down as far as possible. So every digital camera you are likely to pick up at room temperature has SOME noise in its image, same as film. If the noise is controlled down to one part in about 120 your eyes won't detect it, though.

Here are three images that illustrate the idea:

One image has no noise. It's a flat, straight, undeviating 50% gray. Another has 1% noise in it, meaning that some scattering of pixels are ±0.5% above or below the 50% gray. The last (where is it?) has 3% noise in it--a scattering of pixels that run the range of 48.5% to 51.5% gray in them.

Which is which?

Depending on the accuracy of your monitor, you might not be able to tell them apart, but lift them off this page, put them into Photoshop, and you probably can test them by contrasting them up to see the subtle differences. Or just open each in a new page with your browser and read its name.

ANSWER at the bottom of this article.

So a little noise is not so bad, but a lot of noise can make your pictures look like they were processed in the One-Hour Photo place in downtown Elbonia.

How much noise is too much noise? That depends on several things.

1. How big are the noise elements? If you were to look at the above squares from across the room, they probably all look dead equal. Up close, you probably can spot the one with the 3% noise quite easily. I certainly can on my laptop.

So one factor of noise is this: How big are the chunks that contribute to it? Big chunks = Bad. Small chunks = Good. If each pixel is a noise element, you can get away with a lot. After all, that 3% patch is noisy 1 part in 33.3, but they're distributed in single pixel sized differences.

2. Corollary to this is how big is the whole image reproduced? If you shoot a noisy image and shrink it down to Internet scale, the down-sampling of the noise averages it out with its neighbors, to a marked degree, especially if the noise is comprised of single pixel differences (it isn't always).

Want proof? Of course you do. Here are two images. Both shot with the "noisy" 828, but each is a cropped slice from a 816 pixel wide reduction (25% of original). One was shot at ISO 800, the other at ISO 64. Which is which?


Answer at the bottom of the article.

If you printed the full image as a 5x7 print, you would see minimal noise, depending on the subject. But the detail and color would be great.

3. Some of noise depends on its point of reference. Copy each of the three patches above onto a single graphic layout and print them on the world's finest printer with the world's finest glossy paper and most sparklingly perfect ink and tell me if you can spot the difference.

You probably will see LESS of a distinction between them on a print than you see here on your computer screen. But then, there are computer screens that have a somewhat random noise to their phosphors or LCD elements, and those could mask the look of the gray squares, too.

4. Some of noise depends on its local environment. Detail masks noise. You can detect noise in broad areas of continuous tonality much easier than you can in areas of strong small-scale tonal variation.

Blue skies are continuous, grass and foliage is not. Man made graphics are often noise-revealing due to smooth tones of paint or color while images of the Grand Canyon on a clear day may hide noise among the rocks and strata. Out of focus backgrounds will reveal it while your portrait subject's hair will tend to hide it.

5. Some of noise's impact on an image depends on the relationship between the noise and the eye of the beholder. If you can see it and it is disruptive, it's too much. But if it isn't disruptive, your eye may treat it as some other characteristic.

The world of photography is filled with examples of "noisy" images shot backstage, under a street lamp, deep undersea, of a beautiful nude, in available light or of a graphically strong subject that carry the impression of Gritty Documentary, Interesting Texture, or Graphic Counterpoint instead of simply "noisy image".

6. Some of noise's impact depends on how much camera sharpening is present. Punch the image up into the Sharpening + zone, and up goes the noise. Dial in Sharpening - and down it comes. To a degree.

When shooting images at smaller file sizes, the down-sampling idea (Item 2 above) improves detail while tending to blend small scale noise. Lifting the Sharpening setting tends to not recover noise as much. You can get away with less in-camera digital sharpening on smaller files.

7. Some of noise is perceived through chrominance. Black and white images get away with more than color images do. Take a picture with noise in it and drain the color out of the shot so the tonalities are what your eye is responding to and the monochrome image will tend to look like high speed film instead of a noisy color shot. Of course color film gets noisier the higher its ISO goes, but B&W is more forgiving.

So, how does the chip performance of the V1 compare to the 828?

The 828 has an image size advantage over any 5-megapixel image. It's 27.5% wider than the V1's array of pixels, so shooting an image that puts the same subject onto the same number of pixels is not an exact science. But that is what has to be done to make a direct pixel for pixel comparison of the two cameras.

I managed to do it, fairly closely, but not exactly. Still you can see several interesting things in these two examples. What you see here are 100% crops right out of the center of a string of tripod shots that were made a few minutes apart of the same subject. Notice the shadow wander between each series.

If the display on your monitor is a typical 90 ppi, the V1 image would be 28.8 inches wide! And the 828 image would be 36.3 inches wide!

The V1 is on the left. (If these show upon your browser screen one above the other, widen the window till they fit side by side.

They're QuickTime animations, so you can stop their motion by clicking the lower left animation box to pause it. Then you can move the time button back and forth to access the ISO frame you wish to compare to its neighbor.


Overall, the Real Color setting of the 828 has produced a MUCH more accurate color rendition of the scene. Tonal cues are better and exposure of each camera is within 1/3 stop of the other. The 828 image seems a tad darker but the white ice cream truck is not as blown out. No correction has been added to either series. Each series is internally consistent exposure.

Notice that at the lowest ISO settings for each camera (100 vs 64), the detail per pixel of the 828 is slightly greater. Sony is getting more out of each CFA group than it did with the V1. The tree's leaf structure shows this. Noise level is about equal.

By ISO 100 on the 828, a small, rather insignificant amount of noise is showing up. It's a few db more than that of the V1, but nobody would spot it in a print, and you would miss it here if the image were reduced in size. That print would be quite a bit bigger, or more finely detailed overall, if it were made at the same size as the one from the V1.

By ISO 200 on each, a trend has been set. Certainly the 828 image has collected more noise, but so has the V1 to a visible, but still minimal degree. You can shoot either camera at this ISO and get great shots.

At ISO 400 on each, both acquire more noise as expected, but the 828, now five ISO choices higher than its lowest, is not doing as well as the V1. The 828 image is beginning to look grainy while the V1 image is still looking okay. It looks somewhat like a higher speed film image and while the grain is noticed at this scale, a print might mask it to a pronounced degree.

Hitting the ISO 800 setting on each camera introduces quite a bit more noise on the 828 than on the V1. And the noise has shown its true character. It's more "clumpy" on the 828 image, making this setting useful for either the grittiest Documentary look or for conversion to B&W. Or save ISO 800 for your 1M or VGA images where numbered factors #2, 3 and 4 (above) might help.

ANSWER 1: Gray Noise Test. Left is the 1%, middle is the 3% and the right is 0% noise.

ANSWER 2: Which ISO Test. The one on the left started life as ISO 800. It's the same frame from the 828's animation. Meaning that you can shoot with the 828 at 1M or VGA size files and likely never see the noise in the picture at ANY ISO.

More will be here as the Sony Zone pages evolve.


PS: As experience grows with these cameras, so will this review and the Sony eBook. Available now on the order page.

© 2004 Peter iNova. All rights reserved. Do not replicate or link to images without permission. All photos by Peter iNova unless attributed to others. Photos are the copyright of their originators.