I need to work with images that are taken in uncontrolled lighting conditions, that is to say, the images are taken by random users who can't or don't control the luminance. I need a colour space which best copes with it, and provides fairly better results. RGB produces very bad results. I have tried HSV colour space, which is a bit better than RGB, still inefficient. How would 3-channel YUV420 fare in this case? Any help would be appreciated.
It depends on the application, but I've used CIELab before and had good results
YCbCR is generally regarded as the most preferred color space in uncontrolled lighting conditions. Compact color structures can be obtained under a wide range of varying lighting variations under this color space. Work done by Garcia et al [1] (reference attached) shows that YCBCr is more compact than HSV and the later color space is more sensitive to lighting variations.
I have personally used YCbCr over various background subtraction techniques have found it to be better that most other color spaces.
Article Face Detection Using Quantized Skin Color Regions Merging an...
Thanks a lot everyone. Talking about the application, I need to extract some part of the image which is very faint and is almost similar to the background. So, I need a colour space which can perform the background subtraction part pretty well.
To work on lighting L (luminosity) (LAB) space would provide better results. L space is directly related towards the Intensity components. Are you trying for a normalized color space to work further ? If so have a reference color space represents each channels - gray world theory or just use the white space (255,255,255) where you get the normalized image. Another method-Separate the foreground and background through morphological operation and then normalize. But if you mention more clear way - if anything strikes me - would help you. The above two methods I used for medical images and for object detection.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing