Originally Posted by Matthew Gilley
All reviews have some bias in them, of course. Read Ken Rockwell's reviews, for example.
Can't the 7D be tested at these apertures? If it's not sharp at these apertures, based on the DLA, shouldn't that be a concern for anyone shelling out $1700 for the 7D?
That's a good point. On the other hand, looking at Bryan's iso12233 crop chart of the 200 f/2 with the 7D, the 7D looks soft, even with the lens wide open (compare to 1DIII). At first I thought this was just because the pixel density was high enough to reveal weakness in the lens itself. But maybe it is the camera.
Gee, Ken Rockwell biased? Nah... []
I've got the 7d and the 17-55 lens. The pictures originally came out ultra soft but that was because I had been only shooting in stupidly low light. When I shot on a nice day, not too much sun, some cloud coverage, I got tack sharp images handheld, even with my aperture wide open at 2.8! Hope this helps?
Sure, thanks, Jlau. I usually don't put much weight in the odd review that says a camera has soft images, because usually they just did something wrong. But these guys sounded thorough. Nice to hear a report to the contrary.
Sensor resolution is affected by only two things:
- Optical Low Pass Filter (OLPF, or anti-alias filter)
- Megapixels.
Everything else is processing. If the 7D truly is softer (I have yet to see for myself), it can only be for one reason: Canon improved the OLPF. The 50D OLPF was too weak, and allowed a lot of aliasing artifacts. Many photographers love aliasing artifacts (they call it "sharpness"), and even advocate that the filter be removed. Personally, I think the 50D OLPF was far too weak and I prefer the 20D-style softening.
Apparently I don't understand. I thought the 50D has a weak olpf because it has a gapless sensor, and thus no aliasing artifacts except those caused by the bayer filter (far less severe than gap aliasing, I would have thought). Am I wrong?
Microlenses do reduce aliasing, but a Siemens Star comparison of the 20D and 50D still shows far more aliasing in the 50D (i.e. relative to Nyquist or pixel pitch, not relative to absolute spatial frequency). There is quite a bit of luma (non-Bayer) aliasing in addition to the chroma aliasing (which could be said to be caused by the Bayer filter).
Okay. I don't understand how that can be, but then that isn't surprising. I don't understand signal processing at all.
And here is another thing I don't understand, come to think of it: one can do a low pass filter in post processing, so why does one need an optical one?
Wikipedia has a pretty good article on it. The difference between a sensor with poor optical fill factor (no microlenses) and excellent optical fill factor (gapless microlenses) is like the difference between point sampling and box sampling. A point filter causes more aliasing than a box filter, but the box filter still has quite a bit of aliasing.
One way to look at it is from the point of view of MTF. In order to suppress aliasing, the MTF at the Nyquist frequency has to be below 10% (ideally 0%, but most people can live with a little aliasing at such low contrast).
To calculate the MTF of just the sensor at any given spatial frequency, the general
formula is (sin(pi*a*f))/(pi*a*f), where a is pixel width in microns
and f is the spatial frequency in cycles/mm (a black line next to white constitutes one cycle). The Nyquist limit for the
sensor is the spatial frequency of f=1/(2*p).
To then calculate the MTF at the Nyquist frequency for the simple case of 100% fill factor
(a=p), it comes to (sin(pi*a/(2*a)))/(pi*a/(2*a)). Which is the same as
(sin(pi/2))/((pi/2). The end result is 64% MTF at Nyquist. That is far higher than 10%, and will alias badly. But if you reduce the fill factor below 100%, the MTF increases, and aliasing gets worse and worse.
The OLPF reduces contrast at the Nyquist frequency to reduce aliasing.
Because by then it's too late. A 12.8 MP file from a full frame camera should only contain detail up to a certain spatial frequency: 40 lp/mm (using 3 pixels per LP). If you don't have an optical filter, then it will actually contain details that are supposed to be too small for it to see, such as 60 lp/mm. But once they are in the file, there's no way to know which details are too fine and which are normal.
But let's say you have an ideal OLPF. Then the file will contain details at 40 lp/mm and below: nothing above that whatsoever.
Now what happens when you downsample? If you use a point filter and downsample to 3.6 MP, you keep the 40 lp/mm details, even though the smaller file is not supposed to have them. Look at this image, for example:
See the dots of stubble on his face? You're not supposed to see that! The stubble is so small that all you should see is a slight gray blur, not individual hairs. The reason you can see them (as you know) is that I did not use a software low pass filter.
Now here is the same image, but with proper resampling:
It's no longer possible to see individual stubble in sharp detail, but the gray blur is there.
So you are right that we can use a low pass filter in software, but only over the frequencies that are *already* in the image file itself. The problem is that unless you have an optical filter, the file will contain frequencies that are too high to begin with.
Of course, it all makes sense now since reading that last post Daniel! You point the front of the long tubey thing at whatever you want to take a picture of and press the button on top of the black square box. [:P]
http://www.flickr.com/photos/ben_taylor_au/ www.methodicallymuddled.wordpress.com
Canon 5D Mark III | Canon 5D Mark II | Samyang 14mm f/2.8 | Canon 35mm f/1.4L USM | Sigma 85mm f/1.4 EX DG HSM |Canon 70-200mm f/2.8L IS II |Canon 2 x Teleconverter III | Canon 580 EX II Speedlite | Really Right Stuff TVC 34L | Really Right Stuff BH55 LR | Gorillapod Focus | Really Right Stuff BH 30
Posting Permissions
Reply With Quote
