24mm is an interesting focal length for both full frame and APS-C (crop sensor) cameras. On a full frame camera it's a very useful wideangle, maybe the widest focal length that doesn't obviously give rise to images with "perspective distortion". I quote "perspective distortion", because the ultra-wide look is not really distortion in an optical sense, it's more of an unfamilar viewpoint.
On a cropped sensor camera such as a Rebel XSi or EOS 50D, a 24mm lens gives you the same field of view as a 38.4mm lens would on a full frame camera. This is a useful "normal-wide" focal length. If you define a "normal" lens as one with the focal length equal to the image diameter, then a 24mm lens is very close to "normal" on an APS-C sensor camera, which has a sensor diagonal of 26.8mm. The focal length range from 35-40mm on a full frame camera was very popular with rangefinder cameras in the 1950s and 1960s and a 24mm lens on a 50D or XSi gives you a similar view.
Another reason I decided to look at 24mm lenses is that I seem to have collected a lot of them, or a least a lot of zooms which cover 24mm. The lenses that I looked at were:
Obviously I didn't include all the Canon lenses that cover 24mm because I don't have access to them right now. They include:
I looked at each lens both wide open and at f8, and I looked at the center (position [1]), the corner of a Canon EOS 5D full frame sensor (position [23) and an area 62.5% of the distance to the corner of the full frame, which is the same as the corner of an APS-C crop sensor (position [2]), as shown below.
I'm not going to fill this page with dozens of test images. Instead I'll make some comments based on the analysis of multiple test shots taken with each lens (both resolution chart and "real world" images) and show a few representative image samples. Remember of course that I just looked at one sample of each lens, and that it's possible that a lens of the same model that you own would not show the same identical optical performance as the one I tested.
In the center of the image all the lenses were pretty good, even wide open. This isn't unexpected as in a properly constructed lens the only aberration affecting the image quality in the center of the image is spherical aberration. Things like distortion, coma, astigmatism, field curvature and chromatic aberration are all "off-axis" aberrations. The lens that fell at the bottom of this test (and all the other tests...) was the Samyang 18-28 zoom. No big surprise. It's not a good lens and in fact is the standard by which I judge a poor lens. It was cheap 10 years ago and I don't think it's even made anymore. The sharpness in the center isn't awful, but contrast is low wide open and not a whole lot better at f8.
All the other lenses showed decent image quality in the center of the frame wide open, though in each case stopping down to f8 resulted in a slight increase in contrast and sharpness. Any of these lenses can produce an image that's sharp enough in the center for at least 11x14 prints.
All of these lenses are designed for use on full frame cameras, so you'd expect reasonably good performance at the corners of the smaller APS-C frame - and in general you'd get it, as long as you don't want to shoot the lens wide open. By f8 all of the lenses (except for the Samyang) are just about acceptable at the corners of the APS-C frame, but as you might expect, the 24/2.8 prime and the 24-105/4L IS USM yield the highest image quality.
This is the test that separates the men from the boys - or the lenses from the dogs! It's a tough test. All the off axis aberrations will be at their maximum in the image corners. Stopping down helps a little, but a lens that's bad wide open isn't going to become good when stopped down. Even the 24/2.8 prime and 24-105/4L IS USM images are a little soft in the very corners of the full frame image if you pixel peep, but they are clearly better than the other lenses that I tested in this group.
Here are just a couple of samples to show relative differences. The first compares the 24/2.8 at f4 with the 22-55/4-5.6 at 24mm and f4. These images have not been corrected in any way for aberrations. A Canon EOS 5D was used and the image reframed each time to locate the region of interest either in the center of the frame, at the position of the corner of an APS-C frame and at the corner of the full frame image. The images are 100% crops.
As you can see, in the center of the image both lenses give a reasonably sharp image, though the 24/2.8 is certainly slightly better as can be seen from the amount of detail recorded in the bushes. Moving out to the APS-C corner position, the image from the 24/2.8 hardly changes. It's still just about as sharp. This means that if the lens was used on a crop sensor camera like the EOS 40D, the image would be pretty sharp across the whole frame. On the other hand you can see that in the 22-55 zoom image there has been considerable softening moving the the APS-C corner position. Out at the very corner of the full frame image, there is some softening of the 24/2.8 image. The 22-55 image in the corner of the full frame is very soft
Stopping down to f8 improves the images from both lenses, but the 24/2.8 prime is still clearly sharper, in fact the 24/2.8 at f4 is still sharper, even in the corners of the full frame image, than the 22-55 is at f8.
So stopping down the 22-55 gives a little more relative improvement over f4 then stopping down the 24/2.8, but only because is was much worse to start with. Of course these are 100% crops and so represent small sections of an image that's maybe 24" x 36" on a typical monitor screen, so this is pixel peeping. In fact the full frame image from the 22-55 would make a pretty acceptable print at 8x10.
Incidentally, the "stretching" of the detail at the corner of the full frame image isn't due to gemometric lens distortion (which causes straight lines to appear curved). It's a consequence of the optics of a wideangle rectilinear lens. Even a perfect lens would show this effect which is due to the way the image is mapped onto the sensor. The wider the lens, the stronger the stretching becomes.
Canon's DPP (Digital Photo Pro) RAW converter can automatically correct certain aberrations (vugnetting, distortion and chromatic aberration) for a number of lenses, including the 24/2.8, 24-105/4L IS USM and 20-35/3.5-4.5 USM (but not the 22-55/4-5.6).
Here's a comparision of the image quality in the corner of the full frame image (100% crops) of the Tamron 24/2.5, Canon 24/2.8 and Canon 20-35/3.5-4.5, all shot at f8. These are 5D shots at ISO 1600 (hence you can see a little noise).
You can see the effect of correcting CA in DPP for the two canon lenses. It helps the 20-35/3.5-4.5 more than the 24/2.8 because the 20-35/3.5-4.5 shows higher levels of CA to start with.
Again remember that these small (200x200 pixel) crops represent a very small section of a larger image. On my system, with a 17" diagonal monitor and using a 1280x1024 display resolution, these crops are approximately equivalent to looking at a corner 2" x 2" section cropped from a 28" x 42" print. So just because something may look a little soft or at least less sharp then you'd like, in more normal sized print you might not see any problem.
So my overall ranking of these lenses would be as follows:
So if I wanted the best bargain (and best image quality) is a 24mm lens for use on a full frame camera, I'd pick the 24/2.8. It's an older lens design with an arc form focus motor, but the optics are good (and it has floating elements for close range correction, so image quality holds up even at close focus). Focus isn't silent and is perhaps slower than a USM version would be, but a wideangle lens like a 24mm doesn't have to move the elements far to focus, so focus is quite fast anyway. The faster 24/1.4L is 4x the price of the 24/2.8. If you wanted a lens for low light or minimum DOF, it's the obvious choice but I doubt most users would see much improvement over the 24/2.8 when stopped down to f8 or smaller apertures for landscape work. The EF 24-105/4L IS USM is also a good choice. The IS function allows the lens to be handheld in light that would need an f1.0 or f1.4 lens without IS. Image quality is pretty good, but distortion is quite high at 24mm. Fortunately this can be digitally corrected using DPP. The downside of the zoom is the cost (close to $1000) and the significantly larger size and weight then the 24/2.8.
For use on an APS-C crop sensor camera (where a 24mm lens is close to "normal" rather than wideangle), even the EF 20-35 and EF 22-55 lenses will give acceptable corner image quality when not shot wide open, though again the 24/2.8 prime would be my first choice, followed closely by the 24-105/4L IS USM.