9 years in the megapixel race - noise or quality

Kattintson ide a magyar változatért

Native resolution

All camera review sites, without exception display image samples in native resolution. It is definitely a logical step, but can be misleading a bit.
Logical, because cameras are mostly used in their native resolutions, since that's where we can expect the most details.
It can be misleading, if we forget that each pixel will shrink in a given print size, when the resolution is raised, or at same print resolution we get larger prints. Viewing images in 1:1 resolution (when one pixel of the image is displayed on 1 pixel of our monitor) is like using a magnifying glass. If we have a larger resolution image, we get a glass with a stronger magnifying effect. If we think about two prints of the same size, it is really unfair to compare them in different magnification. So even if larger resolution cameras can have worse image in 1:1 view, could have a better print than those with smaller resolution (at least until we hit the resolution limits of the printer or the lab).
When I compare two cameras with different native resolutions, I always resize them to have the same resolution (enlarge the smaller to simulate larger prints, and shrink the larger to simulate smaller prints), so the images become a lot more comparable.

But now we should see the images, or to be precise the crops from those images in their native resolution. But don't forget: we go from 3 megapixel to 15 megapixel, so the resolution increases significantly.

If we click through from G1 to G10 all the tabs, we can see how the resolution has grown over time. The same crop size can show smaller and smaller parts of our standard target. While we can see the 500 Forint bill in full hight at G1 images, the 100 Forint coin fills the whole image taken with the G10. This means that the image with larger resolution probably carries more details of the target. This is definitely good for us, but we also know that with a given sensor size (the size of the sensors in the G line has barely changed during the years), we can only increase resolution if we shrink the size of the individual photo diodes. This will result in smaller dynamic range and worse noise/signal ratio. Of course camera manufacturers know the physics, so they not only increase the resolution, but work on other parts of the system as well, so the noise/signal ratio isn't as bad as we could expect from the numbers. Microlenses over the individual pixels become larger, and production become more sophisticated, enabling each pixel to grow during the years.

Going through the different sensitivity stages, we can observe that Canon has introduced noise filtering in the G6. Up till that point we had to deal with significant noise at ISO 400, while from the G6 the image become slightly washed out, but a lot less noisy. First it was weird, but I think we got used to it.
To be able to evaluate the real noise levels, we have developed all RAW images in Lightroom v2.1 as well. Here we had to skip the G7, since it doesn't support the RAW format.
Going through the images concentrating on the noise levels, we can see that the noise didn't increase significantly. There are differences, though. The most significant is the difference between the images of the G6 and G9. While the G6 has noise larger than 1 pixel in the blue channel, the G9 has a cleaner blue channel with smaller noise “particles” (ISO400 RAW).
It is interesting to see how different the noise reduction on the JPEG images at ISO1600 on the G9 and on the G10. Which one you prefer, is up to you, but the colors on the G9 seem to be more saturated.

Of course, we could compare the images in several other ways too, but I think we can state that crops above show significant improvements.