Focus - recompose: Please convince me

Mattel Barbie Photo Designer Digital Camera

09/18/2006 06:40:05 PM · #1

First, read this: //visual-vacations.com/Photography/focus-recompose_sucks.htm

Now... assuming I want the EYES to be the most "in focus" of anything on my subject, I believe this document to be wrong.

The distance from the subject's eyes to the focal plane of the camera will change by, at most, the distance that the camera moves during the tilt-focus-tilt-shoot (a secondary distanced touched on by the author, but not his main point).

Therefore, if I avoid the tilt and use the upper focus point to establish focus - the ONLY CHANGE made is the "camera tilt distance", (switching focus points in the camera without moving the focal plane does not change the distance, so only the camera tilt is involved). Which does not equal the several inches difference that the author maintains.

I believe the author to be wrong. AT LEAST as far as my goal is concerned: that of having the eyes be the primary focus point. If the author wanted the subject's chest in focus, then maybe he has a point, but I'm not ready to buy that yet.

So... you read the article and digest it. Is he wrong? Or have I misunderstood? (do you think you can set me straight?!?)

agenkin

Nikon MF Nikkor 20mm f/3.5 AIS

09/18/2006 07:08:43 PM · #2

I remember reading and discussing this document earlier. I did some maths (it's trivial here). My finding was that, in principle, this is correct, and *does* matter when using very shallow DOF. Since I like to shoot portraits at appertures of around f/2-2.8, I always switch to a focus point, which would allow me to minimize the effect of recomposing.

The problem is even worse because, to focus on the eye with the center focus point, the camera needs to be tilted back on the tripod head (i.e. get a bit farther from the subject).

Nikon D200

Nikon AF-S DX NIKKOR 18-200mm f/3.5-5.6G ED VR II

09/18/2006 07:44:16 PM · #3

I'm not sure about the maths involved, but I make it a point to have the tripod set much higher than in the diagrammed example when I'm doing portraits.

When the subject is looking down at the camera lens, there is a greater chance of accidentally capturing a double chin. I personally like to have the subjects looking level towards the lens or slightly upward, as I think it helps to make the subject look thinner and the eyes pop a bit more.

Have not experienced a problem with backfocusing with this method. On occasion, I'll play with the focus points when using a shallow DOF, but for the most part I don't focus and then recompose.

Of course, perhaps if I did I'd have better challenge entries. :)

Canon EF 24-105mm f/4.0L IS

09/18/2006 08:27:52 PM · #4

Originally posted by agenkin:

I'd like to know what math is involved.

The only thing I can imagine, based on the author's diagram, is some math that would compute an optimum distance which would keep both the eyes *and* the chest in focus. The two are definitely at different distances. So the question would become, "What is the optimal focus distance to get both in focus for a given aperture?" And I _do not_ believe that selecting a different focus point is going to make much of a difference in determining *that* answer.

The thing is, if you pivot your camera as you tilt-focus-recompose, then that "pivot" minimizes the actual distance moved. In fact, if you think about it, if you were able to pivot around the center point of your sensor, then the actual distance moved, even if you tilted NINETY degrees, could never be more than 1/2 of the size of the sensor and will generally be much less (the closer the final composed position places the eyes to the center, the less the total movement of the pivot). Now there's some math to work out. :-)

The thing is ... in practice ... I switch back and forth between horizontal and vertical composition so often, that it would be silly to keep picking different focus points. It would seriously slow me down and make me ineffective in getting shots where timing is crucial.

For the few "millimeters" off that my focus might be, I think I can live with using just the center focus point.

And besides, on the Canon cameras, the center focus point is the most accurate of the focus points. If you pick a different focus point, chances are your focus won't be as good as it could have been anyway.

crayon

Samsung Galaxy S24 Ultra

09/18/2006 09:27:07 PM · #5

I think the "face detection" in some newer cameras probably would have fixed that. On a side note, the off-focus on the model's face will probably be negligable unless your DOF is very shallow. But then again, if your model happens to have huge buldging chest measurements then maybe it's a good thing to first point the AF on the face before composing again.

gaurawa

Canon EOS-5D Mark II

09/18/2006 09:48:30 PM · #6

I agree with what was said in the article about the focus plane and subject distance...
I will try to answer your questions

e.g. subject is 5ft away. center point is at chest. Eyes are 1ft above the chest. If you tilt you camera to get the eyes in center, the distance would be sqrt(5^2 + 1^2) ~ 5.1ft so about 1 inch. If you dof 2-3 inches you are probably ok...

now if the subject was closer, say 2ft, the difference would be

sqrt(5) - 2 = 3inches and with a shallow dof, thats a problem

Nikolai1024

Nikon D300

Nikon AF Zoom-Nikkor 24-85mm f2.8-4.0D

09/18/2006 10:31:03 PM · #7

This crossed my mind but I never looked into that. Here is my attempt at math. This table shows angle of camera tilt when recomposing the shot as well as how much DOF changes. This could be any units. Feet or meters will work fine.

Example: if you shooting from 5 meters away and you tilt your camera by 10 degrees you DOF will move forward by 8cm.

Angle At 5 At 10
1 0.00 0.00
2 0.00 0.00
3 0.01 0.00
4 0.01 0.00
5 0.02 0.00
6 0.03 0.00
7 0.04 0.00
8 0.05 0.00
9 0.06 0.00
10 0.08 0.00
11 0.09 0.00
12 0.11 0.00
13 0.13 0.00
14 0.15 0.00
15 0.18 0.00
16 0.20 0.00
17 0.23 0.00
18 0.26 0.00
19 0.29 0.00
20 0.32 0.00
21 0.36 0.00
22 0.39 0.00
23 0.43 0.00
24 0.47 0.00
25 0.52 0.00
26 0.56 0.00
27 0.61 0.00
28 0.66 0.00
29 0.72 0.00
30 0.77 0.00
31 0.83 0.00
32 0.90 0.00
33 0.96 0.00
34 1.03 0.00
35 1.10 0.00
36 1.18 0.00
37 1.26 0.00
38 1.35 0.00
39 1.43 0.00
40 1.53 0.00
41 1.63 0.00
42 1.73 0.00
43 1.84 0.01
44 1.95 0.01
45 2.07 0.01

This could be completely wrong. I have not done this in years.

Nick

Nikon AF Zoom-Nikkor 24-85mm f2.8-4.0D

09/18/2006 10:33:42 PM · #8

Originally posted by gaurawa:

now if the subject was closer, say 2ft, the difference would besqrt(5) - 2 = 3inches and with a shallow dof, thats a problem

If all you do is pivot your camera, how can the difference from one angle of measurement (to the eyes) be anything more than a fraction of the length of the sensor?

I totally disagree with the diagram in the presentation. What does aiming my camera at the subject's chest have to do with the distance to the eyes? See... I think that's where I actually differ from the author's point of view. I want the eyes to be the main focus. So I focus on the eyes and then recompose. And when I move my camera, the difference in the distance from my initial measurement isn't going to be very much. Certainly not 3 inches ... not without taking a step backward.

edit: "taking a step backward" is an exageration, yes. But you get what I mean. The camera didn't move 5 inches backward simply by pivoting the focal plane.

Message edited by author 2006-09-18 22:35:32.

Nikolai1024

Nikon D300

09/18/2006 10:48:02 PM · #9

Note: Above table is correct as far as DOF but it does not account for change in focal pane movement which moves as you tilt the camera. It's kinnda hard to explain. That math seems beyond me. I may try tomorrow if not busy at work.

Nick

agenkin

Mattel Barbie Photo Designer Digital Camera

Nikon MF Nikkor 20mm f/3.5 AIS

09/18/2006 10:50:25 PM · #10

Originally posted by dwterry:

I totally disagree with the diagram in the presentation. What does aiming my camera at the subject's chest have to do with the distance to the eyes?

There is nothing to disagree about - it is pure geometry.

Suppose you only use the center autofocus point on your camera. Suppose that you focus using the center autofocus point on the subject's eyes. Focusing the lens *only* depends on the actual distance between the sensor plane and the focus plane (which, in this case, passess through the eyes).

Now, having focused on the eyes, you lock the focus, and recompose. Suppose that now the center focus point is aimed at the subject's chest. (1) The subject's chest is closer to the camera than the subject's eyes, and (2) the lens is locked to the focus distance to the subject's eyes. This means that the focus plane has shifted behind the subject's eyes.

Whether the eyes are still in focus depends only on the DOF now. If the DOF is shallow enough, the eyes will be out of focus.

agenkin

Mattel Barbie Photo Designer Digital Camera

Nikon MF Nikkor 20mm f/3.5 AIS

09/18/2006 11:11:42 PM · #11

I think that I see where your confusion is at. You are thinking that, after recomposing, the eyes are still at the same distance from the camera as before recomposing, right?

The problem is that the focal plane is flat and passes perpendicularly to the sensor plane. So, after you recompose, the focal plane has actually shifted farther, than where it was before. The diagram is a little misleading: there should be a third focal plane depicted there, the one that passes through the eyes *before* recomposing, which should be perpendicular to the red ray that connects the camera and the eyes.

Let's put the diagram into a forumla. Suppose your camera is L mm away from the subject's chest, and M mm away from the subject's eyes. Suppose that there is N mm between the subject's chest and the subject's eyes.

Now, plugging in some numbers. The subject's chest is 1000mm (1m) away from the camera, the subject's eyes are 300mm (30cm) away from the middle of the chest, so the subject's eyes should be sqrt(1000^2 + 300^2) mm away from the camera, that is 1044mm. This means that, after you recompose, the lens is still focused for 1044mm. But you want the focal plane to pass at 1000mm. So, if the DOF is shallower than 4.4cm, you have a focusing problem.

This assumes that the subject's eyes are directly above the center of the subject's chest. This is a simplification, of course, but a believable one.

09/18/2006 11:31:24 PM · #12

Originally posted by agenkin:

Now, having focused on the eyes, you lock the focus, and recompose. Suppose that now the center focus point is aimed at the subject's chest. (1) The subject's chest is closer to the camera than the subject's eyes, and (2) the lens is locked to the focus distance to the subject's eyes. This means that the focus plane has shifted behind the subject's eyes.

I agree with #1, yes, the chest is closer than the eyes. Simple geometry there.

The eyes, however, are still in the same position. And the camera has only moved slightly (the amount necessary to pivot the camera) which has nothing to do with either geometry or the position of the subject's chest.

09/18/2006 11:33:36 PM · #13

Originally posted by agenkin:

Now, plugging in some numbers. The subject's chest is 1000mm (1m) away from the camera, the subject's eyes are 300mm (30cm) away from the middle of the chest, so the subject's eyes should be sqrt(1000^2 + 300^2) mm away from the camera, that is 1044mm. This means that, after you recompose, the lens is still focused for 1044mm. But you want the focal plane to pass at 1000mm. So, if the DOF is shallower than 4.4cm, you have a focusing problem.

Wow... 44mm change in distance around a pivot point of a sensor only 35mm long. That's amazing. I have trouble believing the math. :-)

09/18/2006 11:37:15 PM · #14

I'm still not convinced ... and I had grand plans of coming home from work to set up a test for myself (a real world test with a camera and a wall with markings on it). Unfortunately, family duties have prevailed and I haven't had time to set up the test.

If anyone wants to jump in and beat me to it... I may get to it tomorrow.

You've seen focus charts, like this one. What I'd like to see is a test that measures the focus accuracy before and after a lock and recompose.

fotomann_forever

09/18/2006 11:39:09 PM · #15

Too much freakin' math... simply use enough DoF to lock focus on the eys and yet have the tips of the boobs in focus. :-)

Some things just don't need to have THAT much computation. :-) LOL