How can we avoid interpolating across edges, when we have to compute the current pixel from the neighbors?
ericyche
I can imagine creating the multithreading code or ASIC design for this must be really satisfying because you're able to visually see how the interpolation affects the color and accuracy of the picture.
waleedlatif1
As a result of the interpolation or nearest neighbor implantation of demosaicking algorithms, It makes sense why 2/3 of the data is made up and I assume in the general case this is fine, but when there is high contrast in certain areas, doesn't this cause problems like producing the wrong color? My guess would be that to overcome this there could be different demosaicking algorithms for different types of images like low-light or high contrast.
ld184
I'm guessing that the perceptive field is only so much. By linearly interpolating we are kind of upsampling an image, so long as the person who's looking at the image interprets the observed frequencies as desired, the extra noise caused by linearly interpolation and artificial values would be ignored. I do wonder how much can we make up before this is a concern. On a related note, Nvidia's new video upsampling feature also makes up data...
LeslieTrue
Question about the color distribution of mosaick. Why grids of green color are more than red and blue?
Staffmcallisterdavid
@LeslieTrue there's a good explanation here https://en.wikipedia.org/wiki/Bayer_filter. It has to do with human color vision!
JustinHuey1
What are the other types of algorithms used besides bilinear interpolation and is there a way to gauge whether the output is "correct"?
andrewhuang56
I can imagine that since green is being sampled twice as often, then the green pixel values are subject to less variance. Does this have to do with why we choose the most sampled color to be the one that we are most sensitive to? Also, does this decreased noise in green ever become apparent in low-resolution photos, making the image look a little off?
How can we avoid interpolating across edges, when we have to compute the current pixel from the neighbors?
I can imagine creating the multithreading code or ASIC design for this must be really satisfying because you're able to visually see how the interpolation affects the color and accuracy of the picture.
As a result of the interpolation or nearest neighbor implantation of demosaicking algorithms, It makes sense why 2/3 of the data is made up and I assume in the general case this is fine, but when there is high contrast in certain areas, doesn't this cause problems like producing the wrong color? My guess would be that to overcome this there could be different demosaicking algorithms for different types of images like low-light or high contrast.
I'm guessing that the perceptive field is only so much. By linearly interpolating we are kind of upsampling an image, so long as the person who's looking at the image interprets the observed frequencies as desired, the extra noise caused by linearly interpolation and artificial values would be ignored. I do wonder how much can we make up before this is a concern. On a related note, Nvidia's new video upsampling feature also makes up data...
Question about the color distribution of mosaick. Why grids of green color are more than red and blue?
@LeslieTrue there's a good explanation here https://en.wikipedia.org/wiki/Bayer_filter. It has to do with human color vision!
What are the other types of algorithms used besides bilinear interpolation and is there a way to gauge whether the output is "correct"?
I can imagine that since green is being sampled twice as often, then the green pixel values are subject to less variance. Does this have to do with why we choose the most sampled color to be the one that we are most sensitive to? Also, does this decreased noise in green ever become apparent in low-resolution photos, making the image look a little off?