4
$\begingroup$

I was wondering if any one with experience with both sensors can give some insight on this.

I've heard there can be problems with alignment of the different signals using a 3CCD.

However, when using the Bayer filter, you need to convert, and you have problems of lower resolution, introducing colors that are not there in reality, etc.

Is one of the two best usable? Or is it application specific?

EDIT: Maybe I can add some extra questions to make the question more clear. For the record, I'm not asking this question for a specific project, I'm just interested in the differences between 3CCD or Bayer in general. I know the basic (hardware) differences between 3CCD and Bayer, but I'm looking for general details about the differences.

These type of questions come to my mind:

  • In an application that is based on color will it always add significant color information when using 3CCD instead of Bayer?
  • Are there other solutions instead of 3CCD to overcome the problems with Bayer? (thanks Ktuncer)
  • What exactly are the significant differences between 3CCD and Bayer (not application specific, and if application specific, in what kind of applications could it be significant)?
$\endgroup$
  • $\begingroup$ DSLR manufacturers and their customers seem happy with a single CCD. Nikon's latest has 36MP. $\endgroup$ – Emre May 23 '12 at 6:02
  • $\begingroup$ But DSLR cameras and cameras suited for machine vision are two different worlds.. $\endgroup$ – Geerten May 23 '12 at 6:44
  • $\begingroup$ Machine vision sensors are usually much cheaper. What's your application again? $\endgroup$ – Emre May 23 '12 at 6:49
  • $\begingroup$ My application is in the industrial automation. $\endgroup$ – Geerten May 23 '12 at 6:51
3
$\begingroup$

Generally Bayer mask (also known as CFA = color filter array) imagers are the most popular.

3Chip cameras cost considerably more - you are not only paying for 3x as many chips but the alignment in manufacturing is expensive and the range of cameras is much more limited. The largest are HD (1920x1080) although a lot of cameras actually have much smaller chips and rely on a deliberate half-pixel misalignment to interpolate each colour into a higher resolution - as with a CFA. Since most are aimed at the TV market they are usually video output (so you need a grabber card) and are a fixed TV frame rate

There are potentially aliasing issues with CFA cameras if you have a very small point image which may hit only one of the colour cells, this can be fixed with an anti-alliasing filter which blurs the image slightly, or by software.

One area in which 3chip wins is in low light where you need high colour quality. A 3chip camera detects essentially all the light in the scene in each colour. A CFA will detect only 25% of the Red and Blue and 50% of the green since the other photons are blocked by filters.

$\endgroup$
  • 1
    $\begingroup$ Thanks for your answer. I am especially interested in the key feature difference: color. You say little about, but what is the difference in color quality? Is it significant? Maybe examples where the difference can be of great use? $\endgroup$ – Geerten May 29 '12 at 8:23
  • $\begingroup$ @Geerten which features of the color you are interested in, if you give more details, we can come up with interesting solutions. Such as using two mono cameras with color filters in front of them, where you can eliminate bayer impact and still have a very high quality/low cost machine vision system $\endgroup$ – Ktuncer Jul 8 '12 at 1:59
  • $\begingroup$ @Ktuncer, I have added some more information. Hopefully it's more clear now. $\endgroup$ – Geerten Jul 9 '12 at 6:54
1
+50
$\begingroup$

Since you are asking for other approaches, there's the Foveon X3 sensor used on some Sigma cameras. It uses three stacked layers of sensors, just like color film used 3 layers of emulsion - so you don't have the alignment problems of 3CCD.

http://www.foveon.com/

$\endgroup$
  • $\begingroup$ Thanks. Any experience/information of results using them for image processing/computer vision? $\endgroup$ – Geerten Jul 9 '12 at 11:13
  • 1
    $\begingroup$ When decoding Foveon's RAW image format (raw sensor data), it is still necessary to do some post-processing (e.g. increase color contrast) - I think the RAW format contains metadata with appropriate correction parameters. Good RAW decoders (such as dcraw) does the job. $\endgroup$ – Libor Jul 10 '12 at 13:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.