Larry Carter wrote:
I think to even farther the difference don't use black for the pinwheels rather use a complimentary color or the negative of the colored background.
If you are looking for the worst Bayer result, use red on blue. Which is really no different than review sites knowingly using B&W to appease their larger Bayer sponsers
missionphoto wrote:
If you are looking for the worst Bayer result, use red on blue. Which is really no different than review sites knowingly using B&W to appease their larger Bayer sponsers
These are nice but very artificial examples. If you take almost any real life picture into the Lab color space you will see that most detail is in the luminosity channel (B&W). In fact, there is so little detail in the color channels that they could be blurred severely without substantial effect on the image.
Missionphoto, thank you for the history. Although I'm not 100% in agreement with what you said, it was useful info !
For the resolution chart performance, I would take B/W over Red/Blue with no hesitation any day I would rather be a color-blind than a blind.
I agree with Alex and also Clubshooter.
B/W or Luminosity resolution is far more important than distinguishing the two adjacent colors on the 2D plane because REAL LIFE objects are NOT printed lines on the 2D plane. 3D objects are recognized by their edges. And the sharp luminosity transition at the edge is what makes the features resolved.Those fine furs on the legs of the fry can be recognised because of the dark or bright edges created by the lighting. In that regard, adding twice more Green than R/Blue was a brilliant idea.
Fundamentally, I'm in violent agreement with you that the same number of photo-site, however you arrange it, would result in the same amount of info. It is just that the strength and weakness are different depending on how you arrange them. I just have been challenging the Foveon-chealeadr's claim that Foveon is inherently better than Bayer. How could it be with the same number of photons received from the same area ?
BTW, My observation is that AA-filter seems to have longer impact range than just NN-adjacent pixel.
I think Larry's simulated Foveon sensor data is flawed. To compare the Foveon to Bayer the Fovean image should be presented as three separate images of differing colors. These colors would best be described as reddish, greenish, and bluish. Furthermore each layer would have a different apparent sharpness, especially at small apertures due to the different size of the sensor elements. Because the different layers in the Foveon sensor do not sharply distinguish between R, G, and B I would guess that the amount of processing necessary to produce a finished Foveon image to be about the same if not more than the Bayer algorithms. The same sort of guessing game as to a pixel's color is done with the Foveon as with the Bayer. The data used to do this guessing is collected in a different and perhaps better way but the color of any individual pixel will only be as accurate as the technique used to profile the sensor, the variation among sensors, and the algorithm used to process the data.
When the Foveon sensor was first announced and the first sample images arrived I thought it would be a superior technology but as I have learned more about it the Foveon technology looks at least as problematic as the Bayer. I agree with Pondria that it looks as though the difference in acuity and sharpness attributed to the Foveon sensor is actually mostly the result of the lack of an AA filter. The Leica DMR shows what is possible with the Bayer sensor when an AA filter is absent and I would say that the DMR images look much better than the supposedly equivalent pixel count Sigma SD10.
I am looking forward to seeing samples from the new Sigma but it's not encouraging that no samples have been made available yet.
gdeliz2 wrote:
I think Larry's simulated Foveon sensor data is flawed. To compare the Foveon to Bayer the Fovean image should be presented as three separate images of differing colors. These colors would best be described as reddish, greenish, and bluish. Furthermore each layer would have a different apparent sharpness, especially at small apertures due to the different size of the sensor elements. Because the different layers in the Foveon sensor do not sharply distinguish between R, G, and B I would guess that the amount of processing necessary to produce a finished Foveon image to be about the same if not more than the Bayer algorithms. The same sort of guessing game as to a pixel's color is done with the Foveon as with the Bayer. The data used to do this guessing is collected in a different and perhaps better way but the color of any individual pixel will only be as accurate as the technique used to profile the sensor, the variation among sensors, and the algorithm used to process the data.
When the Foveon sensor was first announced and the first sample images arrived I thought it would be a superior technology but as I have learned more about it the Foveon technology looks at least as problematic as the Bayer. I agree with Pondria that it looks as though the difference in acuity and sharpness attributed to the Foveon sensor is actually mostly the result of the lack of an AA filter. The Leica DMR shows what is possible with the Bayer sensor when an AA filter is absent and I would say that the DMR images look much better than the supposedly equivalent pixel count Sigma SD10.
I am looking forward to seeing samples from the new Sigma but it's not encouraging that no samples have been made available yet.
The apparent sharpness of Foveon stems from the fact that the image has not been interpolatively upscaled, as recorded by the camera. Bayer RGB exposures are intermingled instead of solid field. In order to keep processing minimal, 2 of 3 RGB values are guessed or estimated at every recorded pixel, but the final step of downsizing the final recording back to the size of each RGB exposure is skipped to save the processor.
So the final result of Bayer interpolation is an image that is recorded at 400% its optical size. Recording the image with many more pixels than optically sensed blurs the recording some, just like upscaling any image using an interpolative upsampling algorthim to estimate inserted pixel color values blurs the new recording. It doesn't matter much, since the printing process will always cram whatever number of recorded pixels into the frame dimensions. That said, monitor viewing is becoming as important as print, so it would be nice to reduce the Bayer recording to its native R+G+B size, as a final step.
Foveon is recorded, by default, at 100% it's optical size (the size of each individual RGB exposure), so there is no process-induced blur. Downsample a Bayer image back to the size of its individual RGB exposures and it becomes sharp as a tack too.
P.S. It is very easy to prove this is occurring. The following 100% crops were generated from 100% lossless workflows then recorded and shown here as lossless png's. Canon 5D on top. Sigma SD9 on bottom. Please read the text inserts for more. Note the degradation in one set but not the other.
Steen Bondo wrote:
I hope we will see a lot of your images as soon as you get your SD14.
And that goes for you too, Grant.
I'm sure you'll see something after I get one, but I really think you'd need to get your hands on a Sigma to really see if it is suitable for your style of shooting or subject matter.
I would rather be a color-blind than a blind.
What an expert!!!!