Wednesday, July 17, 2013

Why camera noise matters

If you set your camera to free run, you might think that every image is identical. If so, you’d be wrong. There’s a certain amount of noise in every image, which means that grayscale values will change slightly from frame to frame.

There are a couple of ways to see this for yourself. If your software displays the grayscale value of an individual pixel, just compare the number over a series of images. Alternatively, try subtracting or differencing consecutive images. The results might surprise you.

A variation of four or five gray values is not unusual; some cameras give more than that. Out of 256 shades, that’s less than 2%, but it could be a problem.

Imagine for example that you’re using a caliper tool to measure the distance between two edges. If you’ve set up the image well the change in gray values over the edge will be maybe from 50 to 200. But noise in the camera could make that change to 55 to 205, or 45 to 195. And remember to double that variation because you’re finding two edges.

If you’re using sub-pixel interpolation that will make your measurement change from image to image, even if everything else is constant. On a tight tolerance, that could mean rejecting good parts, or accepting bad.

So, camera noise can be a problem. How do you deal with it?

What a good subject for another post.

1 comment:

liell plane said...

I normally use as much light and exposure time as possible, before DOF and motion blur becomes an issue. After that, set gain very low

For a fuzzy image, instead of using one calliper to find an edge I would use two opposing, and take the average position