Frame Averaging

In TV images, snow or electronic noise detracts from the picture; in RTR, this random noise can also be present. If the CCD camera is subject to heat, the noise level will rise, detracting from the image. Some CCD cameras are kept at a lower temperature by a special cooling unit. The digital image processor can eliminate electronic noise by a process called frame averaging which is a mathematical process. The image represented by a set of numbers, i.e., is made up of shades of gray, where 0=black and 255=white. Each pixel is, in effect, given a value between zero and 255. True image information stays the same from one frame to the next. If numerical values are averaged from one frame to the next, the bogus value plus the negative effects of random noise are diminished. The data in the tables below illustrate the effect of frame averaging.

Pixel Values of 1st Frame
555512
555555
553030
553068
Pixel Values of 2nd Frame
555555
558355
553030
1033030
Pixel Values of Averaged
555534
557055
553030
793049

If frame averaging is used, the image is improved; and the more frames averaged, the better the results for noise decreases with the square root of the number of frames averaged. That is, if 16 frames are averaged, noise reduction would be a factor of 4. There is a limit: averaging takes time, namely, 1/30 of a second for each frame averaged. After a certain number of frames are averaged, it is difficult to see any advantage in increasing the number of frames, and inspection slows.

There are three types of averaging used: integration averaging, averaging, and recursive averaging. Integration averaging requires a still image. Recursive averaging has the advantage that it can be performed on the fly. Number of frames that can be averaged is usually 32 or fewer; with integration and averaging, greater numbers of frames can be dealt with.