Are there any known methods to use existing keypoint detectors (which generally work on grayscale images) on color images? of course, one could always add the results of the detector on all three channels to get $3N$ keypoints, but I was wondering if there exist other established methods.

Specifically, I am looking to use a FAST like detector.

Note, that I am not looking for a color based descriptor, but rather for a color based detector.


Short answer is no, there are none, or maybe none that made it until becoming a de facto standard like FAST or SIFT. This is still an open area of research by the way if you plan to work in this field.

Mostly, the issues are linked to the fragility of color representation in digital images. The common RGB color space is not really relevant (distances between colors in RGB do not match the perceptual intuition of color distance), so you probably have to go Lab or YCbCr first, then work on the luminance channel. Also, be aware of the white balance that can vary a lot between different cameras, or with a given camera but varying light conditions (sunlight, artificial light...).

Furthermore, in most cases the detection and/or description have good reasons to be color agnostic. For example, when you watch a black-and-white movie, your brain can still recognize objects although they have lost their color.


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.