Bayer Sensor - Do we get a quarter resolution?


Status
Not open for further replies.

twnll

New Member
Mar 23, 2008
215
0
0
North
Sorry if this sounds a bit ignorant.

A Bayer sensor have 4 pixels to collect light. I suppose it will then output as one pixel. So in that sense whatever the resolution of the sensor should be one quarter of what is claimed?

eg a 12 MP sensor actually is a 3 MP sensor?

Please correct me if I am wrong.
 

Sorry if this sounds a bit ignorant.

A Bayer sensor have 4 pixels to collect light. I suppose it will then output as one pixel. So in that sense whatever the resolution of the sensor should be one quarter of what is claimed?

eg a 12 MP sensor actually is a 3 MP sensor?

Please correct me if I am wrong.

A bayer array sensor is lined with 50 percent green, 25 percent blue and 25 percent red in a GRGB layout. What this means is that in a 12mp sensor, it has 6mp of green and 3mp of the respectively R & B photosite. So technically you are not getting a quarter resolution sensor. The reason this is so is due to naturally occurring colors in everyday life and our eyes' sensitivity to the green spectrum.

All sensors using the bayer mosaic interpolate the remaining missing pixels by comparing the grgb arrays in the neighboring sensors to give an approximate color.

At 2mp those days, the image quality was low (by todays standard), but still acceptable in reproducing a good image.

That was when fuji came out with their diagonal array and their sensors were one of the best then.. in fact when other brands were rolling out 2-3mp cameras, fuji was pushing 6mp interpolated output from 3mp sensors. I had their 6900z then and it's output still beat those from a 5mp effective sensors.. Fuji interpolation technology has evolved to become one of the most accurate color reproduction cameras with unbeatable dynamic range today.

By current standards though, sensors has become better and better in capturing light that the bayer array processing algorithm has been improved and I can almost say perfected. I say almost because if you were to take a picture of varying pure reds or blue shades, you'll notice some loss in the color range & data when viewing the green and blue channels. Pushing these photos in post will display some loss of details. This is where the larger MF/LF digital backs excel with their large sensors and lower pixel density.

In your typical day to day shoot though, whether using a triple layer faveon or bayer array, the image yielded will be almost similar in quality. The limiting factor would still be the signal to noise ratio as electronics becomes faster and complicated and pixel density increases..
 

Last edited:
Sorry if this sounds a bit ignorant.

A Bayer sensor have 4 pixels to collect light. I suppose it will then output as one pixel. So in that sense whatever the resolution of the sensor should be one quarter of what is claimed?

eg a 12 MP sensor actually is a 3 MP sensor?

Please correct me if I am wrong.

No. The luminance information is collected on a pixel level, so 12 MP is still 12MP. It's the colour information that is interpolated and the demosaicing algorithm is important.

http://en.wikipedia.org/wiki/Demosaicing
 

No. The luminance information is collected on a pixel level, so 12 MP is still 12MP.

For general photography then the Bayer interpolation algorithms do a decent job. But not a perfect job because there is for example great difference between the quality you get from a scanning back or even a Foveon X3 chip.

But if you're shooting something that is in predominantly one of the Bayer colours then you will be greatly affected through increased noise and/or reduced detail/resolution.
 

For general photography then the Bayer interpolation algorithms do a decent job. But not a perfect job because there is for example great difference between the quality you get from a scanning back or even a Foveon X3 chip.

But if you're shooting something that is in predominantly one of the Bayer colours then you will be greatly affected through increased noise and/or reduced detail/resolution.

Unfortunately, this is true, but with Foveon, it is not quite possible to achieve the kind of luminance resolution with the same pixel density. Knowing that the human eye sees luminance with a greater resolution than colour information helps in formulating the Bayer pattern and demosaicing algorithm and other visual/video encoding algorithms where colour information is recorded with a lower resolution.

But to answer TS's question correctly, when it comes to luminance information, 12MP is still 12MP. Foveon if you count the number of photosites, then the resolution is divided by 3. Then people will question why Foveon can't go 12MP etc.. which is not being fair to the technology itself. It has a lot to do with whether the colour information is co-sited or not.
 

Last edited:
Yeap, each pixel of Foveon can capture all 3 colours but for bayer, only 1.
 

Unfortunately, this is true, but with Foveon, it is not quite possible to achieve the kind of luminance resolution with the same pixel density. Knowing that the human eye sees luminance with a greater resolution than colour information helps in formulating the Bayer pattern and demosaicing algorithm and other visual/video encoding algorithms where colour information is recorded with a lower resolution.

But to answer TS's question correctly, when it comes to luminance information, 12MP is still 12MP. Foveon if you count the number of photosites, then the resolution is divided by 3. Then people will question why Foveon can't go 12MP etc.. which is not being fair to the technology itself. It has a lot to do with whether the colour information is co-sited or not.

Hold on. I am looking at things at a pixel level. Personally for starters I've always counted Foveon's resolution at pixel level. To be fair to them I think they have as well; it was only Sigma that started the whole, we actually have 12 megapixels from a 4 megapixel CCD marketing spiel.

When I say that you get a great difference from a scanning back or a Foveon chip then I mean 12 million proper pixels not 4 million x 3.

From a group of say 12 pixels, the Foveon sensor will offer you better quality than 12 pixels from a Bayer array.

And with regards to luminance information, 12MP is 12MP yes but in certain situations there is a lot of useless information. Like I said, with a red subject for example, then if your subject is predominantly red then a lot of those Bayer pixels will not be contributing much at all.

Luminance is all well and good but there can be edging and detail that isn't betrayed by luminance values.

The original question is, is the resolution of a Bayer sensor 1/4 the stated resolution. The answer is no, but it is not the full resolution either (unlike, say, a 12mp Foveon sensor at true 12mp not the marketing spiel, if it existed). And in certain conditions for example when working with subjects predominantly of one of the Bayer colours, you suffer more.
 

No. A 12MP sensor is a 12MP sensor.

A Bayer sensor uses 4 pixels to calculate the value of 1 pixel in the image, correct. But the reverse is also true, i.e. each pixel of the sensor is also involved in the calculation of the value of 4 pixels in the image. (The actual ratio may not be 4 : 1 but this is just an illustration of the principals)


A Bayer sensor have 4 pixels to collect light. I suppose it will then output as one pixel. So in that sense whatever the resolution of the sensor should be one quarter of what is claimed?

eg a 12 MP sensor actually is a 3 MP sensor?
 

Hold on. I am looking at things at a pixel level. Personally for starters I've always counted Foveon's resolution at pixel level. To be fair to them I think they have as well; it was only Sigma that started the whole, we actually have 12 megapixels from a 4 megapixel CCD marketing spiel.

When I say that you get a great difference from a scanning back or a Foveon chip then I mean 12 million proper pixels not 4 million x 3.

From a group of say 12 pixels, the Foveon sensor will offer you better quality than 12 pixels from a Bayer array.

And with regards to luminance information, 12MP is 12MP yes but in certain situations there is a lot of useless information. Like I said, with a red subject for example, then if your subject is predominantly red then a lot of those Bayer pixels will not be contributing much at all.

Luminance is all well and good but there can be edging and detail that isn't betrayed by luminance values.

The original question is, is the resolution of a Bayer sensor 1/4 the stated resolution. The answer is no, but it is not the full resolution either (unlike, say, a 12mp Foveon sensor at true 12mp not the marketing spiel, if it existed). And in certain conditions for example when working with subjects predominantly of one of the Bayer colours, you suffer more.

That's why I'm careful not to use 'pixel' but photosite. Pixel for pixel, the Foveon is definitely going to perform better than Bayer because it already has tri-colour information. Certainly if there is a predominance in a particular colour, then the Bayer would not perform as well. I would certainly hope to see a true 12MP Foveon but that will mean cramping 36 million photosites on the same area, something has got to go. Of course, if they succeed in getting sensors for each colour on 3 different layers like film, then it will be fantastic.
 

Can you recall which are the camera models that are using faveon sensor. From what i read in the book, these sensor have 100%coverage unlike bayer but as the sensors are in layers, they lose out the light senstivity.
 

Can you recall which are the camera models that are using faveon sensor. From what i read in the book, these sensor have 100%coverage unlike bayer but as the sensors are in layers, they lose out the light senstivity.

commercially, Sigma dp1, dp2, sd14, sd10, sd9.. industral cams TOSHIBA TELI & FOMOS CAMERA.. taken from faveon site
 

Yes on the sensor but no on the image. One just need to know that bayers uses interpolation (fill-in) to complete the image. Foveon X3 capture images on sensor native size but will use software extrapolation (expand outward) to increase its size to 13MP if required.

When manufacture claims it is a 15MP sensor, it means there is really 15million photosites on the wafer and the output of the image size is as per claimed AxB pixel size. When you set camera to lower MP, a smaller portion of the sensor is used. For Sigma, the native sensor size is 4.6MP and will produce a image file size of 4.6MP but you can use the software or in-camera processing to expand it to larger megapixel.

Fujifilm uses photo site pairing to increase dynamic range and thus halved their image size but can still extrapolate to a final image if required.

The only advantage in theory (at this moment) on Foveon is that it has more physical wafer space to expand. But as I have understood, Foveon required larger per photo site, thus more MP has its challenges.

The question is which is better is too narrow for the eyes to see on the monitor. Our eyes are more sensitive to light than sensor but the monitors cannot. Neither can we see detail without magnifying it. If it requires magnification means you cannot see with our eyes. In the end, it is up to the camera CPU + sensor to produce the image you want.

Sorry if this sounds a bit ignorant.

A Bayer sensor have 4 pixels to collect light. I suppose it will then output as one pixel. So in that sense whatever the resolution of the sensor should be one quarter of what is claimed?

eg a 12 MP sensor actually is a 3 MP sensor?

Please correct me if I am wrong.
 

Yes on the sensor but no on the image. One just need to know that bayers uses interpolation (fill-in) to complete the image. Foveon X3 capture images on sensor native size but will use software extrapolation (expand outward) to increase its size to 13MP if required.

When manufacture claims it is a 15MP sensor, it means there is really 15million photosites on the wafer and the output of the image size is as per claimed AxB pixel size. When you set camera to lower MP, a smaller portion of the sensor is used. For Sigma, the native sensor size is 4.6MP and will produce a image file size of 4.6MP but you can use the software or in-camera processing to expand it to larger megapixel.

Fujifilm uses photo site pairing to increase dynamic range and thus halved their image size but can still extrapolate to a final image if required.

The only advantage in theory (at this moment) on Foveon is that it has more physical wafer space to expand. But as I have understood, Foveon required larger per photo site, thus more MP has its challenges.

The question is which is better is too narrow for the eyes to see on the monitor. Our eyes are more sensitive to light than sensor but the monitors cannot. Neither can we see detail without magnifying it. If it requires magnification means you cannot see with our eyes. In the end, it is up to the camera CPU + sensor to produce the image you want.

only applies if you are shooting from full frame to crop mode. otherwise, if you choose lower MP, the full area of the sensor is still used. the camera processor will downsize the photo for you.
 

Status
Not open for further replies.