jmac698
16th February 2006, 20:17
It seems likely to me that colors often change on luma boundaries. Typically, there is a 2 pixel lag (or 3.57/6 MHz) of the color change following the luma. Some filters shift the color to the left to fix this "bleeding" effect. I think we should look for a color boundary coincident with a luma boundary some pixels to the left, and if so replace the inbetween color with extrapolation of the right color. This would erase those grey lines you see on the left of composite sources. To be advanced you could "accelerate" phase changes if they repeat, so checkerboards could have some intensified colors.
Anyone up to coding this? Think it's a good idea?
Hmm, even better is a type of colorizing, where if luma at one point represents a known color, then increases in luma , due to the original rgb weighting, most likely mean an increase in green. Does the phase start to shift toward green? Then make it green with luma resolution, otherwise don't touch. Hmm, you could analyze some high quality movies and find the statistics of how the pixels are really related, then work backwards.
Anyone up to coding this? Think it's a good idea?
Hmm, even better is a type of colorizing, where if luma at one point represents a known color, then increases in luma , due to the original rgb weighting, most likely mean an increase in green. Does the phase start to shift toward green? Then make it green with luma resolution, otherwise don't touch. Hmm, you could analyze some high quality movies and find the statistics of how the pixels are really related, then work backwards.