MY TECHNIQUE TO VERIFY THE FOCUSING ACCURACY OF A DIGITAL CAMERA
1. Foreword
Usual focusing targets are misleading for several reasons very well described by Leon Goodman in his site (link to Leon Goodman's focusing analysis). In short, what this gentleman discovered is that: 1) the field of a lens is not planar (well known but neglected in most focusing techniques), and 2) lenses suffer from a "kind of astigmatism" that makes them focus differently in different directions (makes sense but never taken into account before by target builders). These two characteristics, that pest even in the best lenses, make pretty useless all pre-existent focusing targets.
The target of Leon Goodman seems perfect to demonstrate his claims but difficult to use for focus accuracy determination. My method is based on a hybrid technique, partly inherited from the method of Francis Poon (link) but mainly rooted on the findings of Leon Goodman with some of my additions.
This effort results from having used several targets with inconsistent results. Inconsistent between them and with what I observed in my practice (more of this in point 5. Possible Problems).
2. Target
Basis of the method. The target is mounted in a board that slides inside a box. The extreme positions are one for focusing and the other for shooting - the focusing pattern is substituted by the focus evaluation pattern before the real image is taken.
Focusing target. The focusing patch may be any grayscale image with adequate contrast but without a repetitive pattern. One that I borrowed from someone in the net and resized for my purposes can be downloaded HERE. This target is a little too small for shooting with a 50/80 mm from more than 2 m distance. For 200 mm it is ok above 2 m.
Focus evaluation target. The shooting target follows what I have learned from Leon Goodman's target, and consists on ellipses with an aspect ratio 1:1.41 such that, when photographed at 45º they appear as circles. This target is adequate for a 50 mm lens at about 1 m. A second larger shooting target, useful for distances of 2 and 4 m, is also added in the opposite side. Notice that, in the case of this larger target, only the smaller ellipses are aligned with the center of the image. These patterns were done in CorelDraw as vector images and can be downloaded HERE, also a bitmap copy of super B size (360 lpi) can be downloaded HERE. Other details of target construction may be observed in the figures where the red rectangles indicate the focusing patch and test target, and the pink rectangle the large test target. The blue crosses indicate the position of the camera sensor in each configuration. Measurements are in centimeters (x1.41) and the construction does not require excessive precision, since what we are searching for are not millimetric errors in the focusing. The whole construction should be attached (glued or whatever) to a heavy object before use.
3. Method
1) Position the camera at 45º to the target (other details are obvious). Target in Focusing Configuration. Register the target-to-camera distance.
IMPORTANT: (i) The illumination must be even for the application of the measuring method of Leon Goodman; (ii) The image of the circles must have a dimension such that it is perfectly well resolved with the camera resolution, no jagged edges should be seen at 100% magnification. This implies larger circles for larger distances. I have noticed that above a certain dimension the results are coherent, but smaller circles deviate the result to the front focus side. (iii) The focusing target must largely exceed the dimension of the central focusing mark in the viewfinder (if needed change the target by a larger one).
2) Turn on the camera self-timer.
3) Turn the focusing ring to shorter or longer distances (focus depends from which side the lens approaches). Focus the camera half pressing the shutter release button several times until the focus is attained (even with professional cameras the best focus is seldom obtained at the first attempt). If your aim is not to test for maximum focus accuracy but for real world speedy action, focus only once and continue to 4).
4) Fully press the button. With the self-timer on, the focus will be locked (you should verify if this is the case with your particular camera).
5) While waiting for the shutter, change the target to the Shooting Configuration (it goes without mentioning that the target cannot change position in this step).
6) Repeat 2) to 5) at least 3 times. If, after 7), you find there is inconsistency of results verify the technique and repeat.
7) Analyze the sharpness of the circular patterns as a function of distance from zero using the method prescribed by Leon Goodman (Photoshop/Histogram/Standard Deviation).
4. Case study
Camera: Nikon D70, Lens: 28-70 f:2.8 @ 50 mm f:4.0, Distance from target: 80 cm, Focusing from far (infinite).
The error (averaged for focusing from far and near) is of +0.3%, what I think is very good for a D70 (see below). At this distance, the error is the same with all the lenses I tested.
Results with D2x (recently fixed by Nikon from front-focus bias) with both, 28-70 f:2.8 and 80-200 f:2.8, are consistent at all focal lengths and distances (1-4 m) except for the 28-70@35 having a 2.8% back focus that disappears at 50 mm. At all other focals the error is negligible and well within the standard deviation observed (error between -0.3 and +0.3%). The 28-80@50 3.3-5.6G (an inexpensive lens that came with the D70 kit) exhibit a constant back-focus of 2.5%.
I used relatively low light intensities (1/20s f:4.0 @ ISO100). Stronger light or more luminous lenses will mean smaller standard deviation.
5. Possible Problems
1) It is important to tests at different distances. The graphic shown above for the D70 was made with the camera at 80 cm from the target (most procedures recommend short distances). However, at 100 cm already a negligible front-focus is observed, and at 200 cm a clear, reproducible, -3.5 cm (-2%) error is observed. The error is even larger at 400 cm. My previous inconsistent results with several targets may result from this problem. It is certainly a fault of (this ?) D70 since it is not observed with the D2x with the same lenses.
2) From these and previous tests I suspect that the error in focusing also depends on the light temperature. I have obtained different results with day and tungsten light. However, I have not done a systematic study. All the results above are with tungsten.
(The method was first posted in 2006/12/27)