Comments on: When is 4:4:4 not really 4:4:4. http://www.alisterchapman.com/2011/02/11/when-is-444-not-really-444/ DoP and Cinematographer Alister Chapman's Blog Sat, 18 Feb 2012 09:05:35 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: alisterchapman http://www.alisterchapman.com/2011/02/11/when-is-444-not-really-444/#comment-100 alisterchapman Tue, 15 Feb 2011 19:50:47 +0000 http://www.xdcam-user.com/?p=349#comment-100 Yes and no. Unless there are at least 1x B sample, 1x R sample and 1x G sample for every pixel of the final output resolution then R and B with a Bayer sensor are under sampled compared to G. To not have a difference in R,G and B resolution would require a sensor with twice as many H and V pixels as the output resolution so that each output pixel is represented by all 4 pixels from the GB,RG matrix. But this is not what manufacturers are doing. They are using a sensor with for example 4k x 2k pixels and calling that a 4k resolution sensor. You can interpolate between pixels as much as you want, but if there isn't a sample in the area you are representing then that data will be missing and what is actually there can only be estimated, thus any true detail will be lost. Looking at a bayer sensor you have a row of pixels that are GBGBGBGBGBGB and the next line down is RGRGRGRGRGRGRG. So the first line has NO red samples and the following line has NO blue samples, so the B and R vertical resolution is half that of G. Horizontally there is one G pixel for each B and for each R. Luma information is derived mostly from the green pixels (which is why there are more of them) but also uses some information from the R and B. In addition as the G pixels are diagonally adjacent pixel interpolation is more accurate. Because of all this the luma channel can achieve a resolution not far removed from what it would be if it was a monchrome sensor. Color though is far more problematic as the filters are not all that strong and there is a lot of leakage. This leakage alone reduces the effective chroma resolution, on top of that trying to extract accurate colors requires extensive interpolation (in part due to the cross color leakage) with adjacent same color pixels. As these are always at least a full pixel away from each other the resolution is reduced further. The end result is that chroma resolution is significantly lower than luma and accordingly G resolution is higher than R or B. Now IF you use a lot more pixels than your desired final image resolution then this might not be an issue. Raise the sensor resolution so that you achieve full resolution in the B and R channels are you'll be OK, but that means double the H and V pixel count, which is not what camera manufacturers are doing. Yes and no. Unless there are at least 1x B sample, 1x R sample and 1x G sample for every pixel of the final output resolution then R and B with a Bayer sensor are under sampled compared to G. To not have a difference in R,G and B resolution would require a sensor with twice as many H and V pixels as the output resolution so that each output pixel is represented by all 4 pixels from the GB,RG matrix. But this is not what manufacturers are doing. They are using a sensor with for example 4k x 2k pixels and calling that a 4k resolution sensor. You can interpolate between pixels as much as you want, but if there isn’t a sample in the area you are representing then that data will be missing and what is actually there can only be estimated, thus any true detail will be lost.

Looking at a bayer sensor you have a row of pixels that are GBGBGBGBGBGB and the next line down is RGRGRGRGRGRGRG. So the first line has NO red samples and the following line has NO blue samples, so the B and R vertical resolution is half that of G. Horizontally there is one G pixel for each B and for each R. Luma information is derived mostly from the green pixels (which is why there are more of them) but also uses some information from the R and B. In addition as the G pixels are diagonally adjacent pixel interpolation is more accurate. Because of all this the luma channel can achieve a resolution not far removed from what it would be if it was a monchrome sensor. Color though is far more problematic as the filters are not all that strong and there is a lot of leakage. This leakage alone reduces the effective chroma resolution, on top of that trying to extract accurate colors requires extensive interpolation (in part due to the cross color leakage) with adjacent same color pixels. As these are always at least a full pixel away from each other the resolution is reduced further. The end result is that chroma resolution is significantly lower than luma and accordingly G resolution is higher than R or B.

Now IF you use a lot more pixels than your desired final image resolution then this might not be an issue. Raise the sensor resolution so that you achieve full resolution in the B and R channels are you’ll be OK, but that means double the H and V pixel count, which is not what camera manufacturers are doing.

]]> By: Hubert Pietrzykowski http://www.alisterchapman.com/2011/02/11/when-is-444-not-really-444/#comment-99 Hubert Pietrzykowski Tue, 15 Feb 2011 10:53:05 +0000 http://www.xdcam-user.com/?p=349#comment-99 Not true. It all depends on the number of photocells in the sensor and how it relates to the output resolution (raster). The fact that area occupied by R/B is smaller than G only affects the fill factor. Not true. It all depends on the number of photocells in the sensor and how it relates to the output resolution (raster). The fact that area occupied by R/B is smaller than G only affects the fill factor.

]]>