abqnmusa wrote:
If uprezzing had no impact on image quality we could save money and use a 3 MP or 6 MP camera. Just uprez it up to 21 MP. Of course that would not work.
I have obviously not made myself clear. I am not suggesting that a 3 Mpixel camera contains the same amount of pictorial information as a 21 Mpixel camera, nor that a 12 Mpixel camera does. I made the mistake of assuming that was obvious.
What I am suggesting is that if you uprez your 12 Mpixel camera to 21 Mpixels and then print it, you will not be able to distinguish the two files except under a microscope. The same goes for the 3 Mpixel file. A direct print will not appear different from an uprezzed one UNLESS you print so large that you can identify and compare individual pixels. Generally photographers try to avoid printing in a way which resolves the shapes of individual pixels.
thedigitalbean wrote:
Uprezzing with a nearest neighbor filter is guaranteed to do nothing to your original image (all pixels in the original image will be exactly preserved in the uprezzed image).
This is correct only if the linear uprez factor is an integer, i.e. if you turn one pixel into 4, 9, 16, 25 pixels and so on.
The resulting uprezzed image will look like @$$ though.
No - it will look exactly like the original in every possible way. But you have illustrated the psychological effect of viewing the pixel grid and the bias that is introduced in a comparison at different magnifications. You are saying the file looks like @$$ compared to a more magnified copy of itself, which illustrates my point that magnification matters when scrutinising crops.
Perhaps we should compare with compare uprezzed images with a NNB filter instead
I have done this and the difference between the nearest neighbour uprez and the bicubic uprez is negligible in terms of pictorial resolution. The differences between the two files are merely a matter of sharpness, not resolution of image data.
People have got to stop admiring the exquisite squareness of their pixels and learn some arithmetic.
I realize these are Canon samples, but until we get more independent field test reports we have to start somewhere.
To me, notwithstanding its superb video imaging capabilities, this imaging device (camera) is pointed squarely at the action/sports/wildlife photographer.
In a non-photographic context, the differences between NN and bicubic interpolation - yes the correct term - can be quite dramatic depending on the data and for images you would not use NN interpolation at all. Bicubic will in general do a far superior result in filling in the data on the finer grid. I have to use interpolation a lot in my work, and only ever use bicubic interpolation. However, the quality of the results will depend crucially on the level of detail and how rapidly it is changing. This is why some advocate a multistage process, whereby you only uprez a little at a time. Something with coarse detail changing slowly will be far easier to interpolate more accurately. While data with rapidly changing values will be more problematic. In this case if I run my program on a finer grid and compare it to the interpolated data from the coarser grid, it depends whether the coarse grid data was not so coarse as to miss data, which it cannot therefore intepolate. So what I'm saying if the camera with the lower res, has a enough res to resolve fine detail, then using bicubic or more sophisticated, you should do a good job when you uprez this to the same pixel count as the higher res camera. But if the low res camera is too low res to reproduce fine detail, when upressed it will be poor in comparison. I guess a 13MP 5D is good enough to hold up very well to a 21MP 5D II in this test and only at very large prints where even the 5D II is upressed will the differences be great.
Richard are you sure you don't mean bilinear interpolation rather than NN, because for images bilinear will do a much much much better job. You wouldn't use NN.
Here are the definitions used of the methods used in PS.
Nearest Neighbor: The Nearest Neighbor interpolation is the fastest option. However, it is also the least sophisticated. It simply takes the color of a pixel and assigns it to the new pixels that are created from that pixel. Due to this rather simplistic approach, Nearest Neighbor interpolation does not create an anti-aliasing effect. This leads to problems with jaggies. Consequently, Nearest Neighbor interpolation is considered to be incapable of producing photographic quality work.
Bilinear: Bilinear interpolation uses the information from a pixel (let's call it the original pixel) and four of the pixels that touch it to determine the color of the new pixels that are created from the original pixel. Bilinear uses rather simple, linear calculations to do this. The Bilinear interpolation does have an anti-aliasing effect. However, it is not considered good enough for photo quality images.
Bicubic: Bicubic interpolation uses the information from an original pixel and sixteen of the surrounding pixels to determine the color of the new pixels that are created from the original pixel. Bicubic interpolation is a big improvement over the previous two interpolation methods for two reasons: (1) Bicubic interpolation uses data from a larger number of pixels and (2) Bicubic interpolation uses a Bicubic calculation that is more sophisticated than the calculations of the previous interpolation methods. Bicubic interpolation is capable of producing photo quality results and is probably the method most commonly used.
Bicubic also introduces a 15-20% pixel to pixel contrast loss, and is VERY poor at detecting contrast in the 45� directions. This is why you might get aliasing effects from "curved" contrasts, like a phase-plate. It is though one of the "safest" ways to resample a pixel surface. Due to the sampling restriction you shouldn't use it to downsample to less than 25% in one step - the algorithm will start to "loose" information from the original by simply leaving the information in between the sampling areas for the new pixels out in cyberspace.
In Sweden it seems the cameras are ready for delivery on next monday in most venues. I'm looking forward to some hands-on testing... :-)
[edit] - that seems to be a mistake on the distribution part - cameras will be on sale late this month... :-( [end edit]
Why? So that I know which image is from which camera. There seems to be a slight exposure difference (or maybe different in camera tonal response curves). Or, if you ask why I ask if you have printed the results, the answer is because on-screen and on print are two different things.
Thanks for writing it ... (Interesting stuff. My questions are intended for interest rather than criticism, of which I have none.)
brainiac wrote:
But you have illustrated the psychological effect of viewing the pixel grid and the bias that is introduced in a comparison at different magnifications. .
Everybody; print this and stick it to the monitor while pixel-peeping.
) the sunny conditions.