Why Isn't This Photograph Pin Sharp?

Question

I Have Been Trying Out My New Canon EF 24-70mm and I'm pretty underwhelmed by its sharpness.

Here is a RAW file shot from a solid tripod on an (almost) windless day using a remote. Obviously it isn't a prime, but it's a pretty good lens. I can get this into a reasonable state using the sharpening in ACR. I wonder if my expectations of sharpness are unrealistic?

Settings (on a 5D Mk3 ETTR):

70mm ISO 100 1/100th ƒ7.1

The stand of trees was approximately 150m away and I focused at 35 Meters.

Is this picture as sharp as I can hope for? Can anyone take a guess at why it isn't sharp?

Note: I'd be really interested to hear people's opinions. I have not been doing this for long, so I have a limited set of experience to draw on.

Answer 1

It doesn't look too bad to me.

You have to consider that when you're looking at a 5D mkIII image at 100%, that amounts to a considerable enlargement. It's rare to get something really pin sharp at that magnification.

The focus point is quite forward so the trees in the centre of the frame are at or near the far limit of the DOF. That combined with the softish lighting is probably giving the effect you're seeing here.

The far left and right sides look about the same, so no evidence for any decentring, they just look out of focus to me, with a touch of lateral chromatic aberration.

It's important to be aware of field curvature, when making test shots I would live view focus in the centre and then again in each of the four corners and compare the results. That should rule out any DOF / field curvature effects.

Answer 2

The blur can be measured by converting to XYZ colorspace and zooming into a tree trunk with a bright sky as the background. You then measure the brightness profile accross the rapid change in brightness (make sure you pick an area with small gradient in the direction parallel to the tree trunk). I then used this method to estimate the blur.

Since the image is actually quite sharp and already after one pixel into one side you are already quite close to the value the brighness takes well inside the region, you will underestimate the gradient right at the boundary. To deal with this issue, I put g(d) = 1/2/[1+exp(b d)] which has the right asymptotic behavior and fitted b. I find b = approximately 2, which yields 1/4 for the gradient at the boundary and it follows from this that the blur diameter is about 2.5 pixels (gradient = 1/(pi R) where R is the blur radius).

Let's see if this is consistent with focussing with a F = 70 mm focal length lens at a distance of d1 = 35 meters, while the trees are at d2 = 150 meters and the 0.00625 mm distance between two pixels mentioned by Matt Grum above. The distance x1 of the sensor to the lens follows from:

1/x1 + 1/d1 = 1/F ------>

x1 = 70.140281 mm

Had Pedr focussed on the trees, then the sensor would have been at a distance x2 from the lens with x2 satisfying the equation:

1/x2 + 1/d2 = 1/F ----->

x2 = 70.0327 mm

This means that when focussing 35 meters away, the light from the trees reach focus a distance x1-x2 = 0.108 mm in front of the sensor. The aperture is D = 70mm/7.1 = 9.86 mm. This means that the angle between the light rays from the trees that enter the lens on opposite sides, is given by approximately D/x2 = 0.141 radians. Then when these light rays move on from the focus point to the sensor, they travel an additional distance of 0.108 mm, therefore they are then 0.108*0.141 mm = 0.015 mm apart. Dividing by the pixel distance of 0.00625 mm, gives a blur diameter of 2.4 pixels.

The blur is slightly broadened by diffraction; the blur diameter due to diffraction for lambda= 500 nm wavelength light is 2*1.22*lambda/D *x2/0.00625 mm = 1.4 pixels, so due to diffraction a single light ray illuminates slightly more than one pixel.

So, the expected blur is pretty much the same as I estimated from the picture. To accurately sharpen the trees, you could make a more accurate fit of the function g(d) by considering many parts of the image. Instead of approximating the point spread function by a uniform blur (or a Gaussian or any other ad hoc choice), one can then then actually compute the point spread function p(s) that yields precisely the function g(d) using the relation:

p(s) = 2/pi Integral from 0 to infinity of g''[s cosh(t)] dt

Sharpening with deconvolution algorithms using the computed point spread function p(s) should yield much better results than using standard sharpening tools.

Answer 3

I can't look at the RAW image at the moment, but it should be fairly sharp, but likely won't be pixel sharp on a 5D Mark iii at 100% magnification. The 24-70 f/4L isn't a prime lens and it isn't the f/2.8II. It isn't as strong of an optical performer and 22mpix is a lot of image data. (I use the f/2.8 II on my 5D Mark iii regularly.)

The f/2.8II will come pretty close to resolving pixel accurate images at 100% zoom in the center, but it is also much sharper (and any of the decent primes will still beat or very nearly beat the f/2.8 II.) Comparatively, the f/4 can't produce as sharp of images and so some amount of noticeable inaccuracy is to be expected, even when properly focused.

That said, you may have a micro-focus adjustment issue, so looking in to calibrating your autofocus micro-adjustments would be worth reading up on. You can either manually work through something like DotTune or you could try installing ML and use their dot tune tool to help out. It's pretty quick and easy to do.

You might also want to try a shot at a lower aperture. According to DXoMark, the sharpest aperture for that lens is f/4.0, so you may be giving up some sharpness with the smaller aperture.

Answer 4

The answer to your question is; No you’re not being unrealistic to expect more sharpness from this lens. This is an L lens and they do not get any better for such Zoom Lenses. I am giving my opinion based on experience. I have found that a lot of Zoom lenses do not work straight of the box and need tweaking. I have had to tweak at least 90% of my lenses since adopting the digital scene to increase sharpness.

I apologise up front if my answer covers anything you may have already looked at, but for this particular lens, which I also use on occasions, the first thing that I do before taking a single shot; is to check the focusing. I keep it very simple and and use the box that the lens came in, to do a micro-adjust and correctly calibrate. In the past, I always used to send the lens back for calibration but now it is so easy.

The issue is that Camera/Lens Manufacturers produce their Cameras/lenses within a tolerance range and sometimes, it is possible to fall at the extremes of both limits.

For example, say the Lens is +5 out and the Camera is -5, you will be fine and have a perfect focus, likewise, if your camera is +5 but the lens happens to be -5, you still get a sharp focus. you are non the wiser if either of the variables are accurate or not.

The problem comes to light, if the Camera is 100% Accurate and the lens is out. or if the camera is + 5 and the Lens is also +5 or +3 or whatever. This issue causes the lens/Camera combination to either focus slightly before(back Focus) the intended subject or slightly after (front Focus), depending on the plus minus figures.

To correct this, it is a fairly easy job and can be done for each different lens which is auto saved and readjusts itself each time the lenses are switched.

Go into your MENU and find AUTO FOCUS DRIVE and select AF/MICRO ADJUST.

There are many techie ways of doing this, However, I just use the Lens Box. I place it on a sturdy surface with Blutac. Place my Camera on a sturdy Tripod between 10-20 Feet Distance. It can be more, makes no difference.

I make sure that the box is at a slight angle to the camera, about a 30 Degree Turned to the right. In Autofocus mode, I use a single Focus point, press the manual zoom button to get a closer look and take a shot of a letter somewhere in the middle of the line. I then go into the Micro Adjust and increase by +1 and take another shot and I keep doing this until I reach the end. I then reverse the process and go the other way. Generally, about 20 shots. ( you don’t have to take this many, but its just me being anal!)

Once done, it is time to compare each shot on the computer. By having the box at an angle, you will be able to tell if the camera is back/front focusing and be able to determine, which of the 20 shots is the most in focus. Ensuring that you have kept a note of the plus/Minus, that is the setting you need to adjust at.

Sometimes, you may find that the same lens will have slightly different adjust at the different ends of the focal lengths, doesn’t happen that often, but I do like to check that too.

if this doesn’t work, then of course, you also have to consider other variables such as wind. the speed you have used of 100sec may be too slow ( just a thought). The depth of field will also make a difference. you may also consider disabling the Highlight Tone Priority in your camera.

Hope this help, and please do let us know how you get on. Good luck