1
$\begingroup$

I have two cameras, and I'm up to perform some matching stuff on videos captured by these two cameras since I'm a CV guy.

The problem is the two cameras have very different performance in terms of color and brightness. The same object in the same illumination appears different in two cameras.

Is there any academic work on measuring something like the "rgb response curve" of a camera? I want to unify the rgb response of these two cameras to guarantee the same object "looks" the same in two cameras.

Note Below is just an intuitive illustration of my problem. Same object appears different in different cameras.

enter image description here enter image description here

$\endgroup$
2
$\begingroup$

Here you have two options.

1) Color calibrate your cameras (radiometric calibration) to the same reference. Then try matching. There are many academic works on this topic, one of them being: http://research.microsoft.com/en-us/um/people/yasumat/papers/rankcalib_PAMI12_preprint.pdf

2) Use a vignetting invariant (or color invariant) feature descriptor. For example you could try to work with gradients instead of intensities and so on. This is harder, but if you could achieve it you get rid of the cumbersome calibration procedure. There are also valuable works done in this field, such as:

http://www.cvip.louisville.edu/wwwcvip/research/publications/Pub_Pdf/2006_2/CSIFTA%20SIFT%20Descriptor%20with%20Color%20Invariant%20Characteristics.pdf

Good luck,

$\endgroup$
  • $\begingroup$ Thanks for this reply. The first point is exactly what I want. However, 'color invariant' features should be less distinctive than 'color variant' features and since the two cameras are known, we can take the 'free lunch' of harnessing the radiometric calibration information. $\endgroup$ – SolessChong Mar 8 '14 at 6:57
  • $\begingroup$ BTW could you provide any more keywords on your first point? I'm non-native English speaker so the most difficult part for me is to come up with the keywords for my question :) $\endgroup$ – SolessChong Mar 8 '14 at 6:58
  • $\begingroup$ I am also a non-native speaker, but I'll try: dark current, vignetting, Quantum efficiency, radiance, irradiance, exposure, RGB distribution, camera shading, sensitivity, equalization and so on. Generally LAB space might be more suitable as it is the closest space to human vision. You might also consider it. $\endgroup$ – Tolga Birdal Mar 8 '14 at 11:24

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.