Double Vision: Cine Lens Quality
Author: Ian Marshall
Principal Optical Development Engineer
Cine lenses are used to produce movies, TV productions and adverts. Producers are trying to emotionally engage viewers with the content, whereas the imaging quality of the lens obstructs or enhances this engagement. In particular, the aesthetics of a lens are not fully understood from a scientific standpoint. There is ‘double vision’: the users want more than lens designers fully understand. This paper takes data from two lenses tests conducted by users and a sample still image taken from the web to illustrate some of the underlying lens design theory.
About the Author
To introduce myself, I am currently working in aerospace and defence, but have worked as a lens designer for Cooke Optics designing 5/is and miniS4s. I am basing this presentation on that experience and subsequent viewing of sample videos from the web.
This paper is based on the following premise:
Cine lenses are used on movies, TV productions and adverts where the cinematographer is attempting various forms of emotional engagement of the viewers.
The lens and camera system, including post processing and projection, creates a ‘porthole’ through which the viewer engages with the artistic composition.
The imaging quality of the lens obstructs or enhances this engagement. Small variations in lens characteristics (including aberrations) are therefore important to end users.
An example of a quote taken from the web is “I have had discussions with rental companies about lenses looking better than other lenses with greater MTF”.
This paper takes data from two lenses tests conducted by users and a sample still image taken from the web to illustrate some of the lens design theory affecting the tests.
I propose that there is ‘double vision’: the users want more than lens designers fully understand.
The characteristics that the users want are usually compromised or traded off in a conventional lens design. Furthermore, some important characteristics are not understood from a scientific standpoint
To illustrate the way lenses might be used on a set, I have copied a clip from the ‘The Big Ole Lens Test Party’ with permission from Ben Eckstein et al.
In this view the camera is in a single position to capture images through different lenses on the same focus settings; so as to look for different properties in the lenses. By keeping all else the same, the differences between the lenses are exposed.
On a real set, there will usually be more than one camera fitted with different focal length lenses set to different focus distances. Because the images produced by different lenses have to have closely matching aesthetics (else the viewer is distracted by the lens differences); the imaging characteristics of different lenses in a set working at different settings must also be closely matched.
The actors repeat short scenes many times, and the images from different perspectives are carefully edited to avoid mismatch.
The first issue that I am presenting is focusing. The focus puller controls the lens focus to match the actor’s positions using a variety of methods, which require the focus marks on the lens to be accurate.
The clip shows how the story line is controlled by accurate focusing which controls the centre of attention. The focusing is on the actors faces, it follows the storyline, and is accurate and smooth.
Notice also how the centre of attention, and hence the sweet spot required of the aberration compensation, may be quite a long way off centre.
(please double click the separate video clip ‘ZeissCP2-windows.wmv’)
Because the focus puller controls lens focus to actor’s positions, the focus puller does not want the focus distance to vary with aberrations on stopping down. He uses a variety of methods and needs the focus marks on the lens to be accurate. Some methods use the lens focus marks as a ‘ruler’.
Also, the through focus ‘roll off’ needs to be consistent from view to view and lens to lens. This refers to the aesthetic of how the image blurs and changes colour (due to secondary spectrum) as the object changes distance to the lens.
Therefore aberrations affecting focus roll off should be consistent across the prime lens set. And lens aperture settings that vary between views of the same scene must not change the look. Hence variations in aberration form with aperture should be controlled.
The need to match these aberration properties between different focal length lenses in a set is a demanding task. And some users stop lenses down two stops from full aperture as a rule of thumb to reduce variations.
To illustrate the scale of when an aberration type is too large to be acceptable, I have generated a simple lens design with higher order Spherical Aberrations. The transverse aberration curves show about 0.035mm of spherical aberration at full aperture (corresponding to T/1.4). At T/2 the aberration form is mostly vignetted.
A simple calculation of monochromatic MTF through focus shows the variation of peak MTF at different positions at the sensor plane.
The longitudinal difference of 0.017mm is a variation that some users have stated to be a problem in lens blogs. As a general rule, I have noticed that lenses without full fiducial markings on the focus ring suffer from this variation. Therefore I suggest users should beware lenses without such marks.
The second issue I am presenting is the well-known out of focus effect called ‘Bokeh’.
Taking the previous demonstration lens design, one can trace rays through the sensor plane to simulate the effects of going out of focus. As shown in the ray trace diagram, the lens aberrations cause hard or soft edges to out of focus highlights, depending on which side of focus the highlight is. In this diagram, the small ray patterns might represent the focus roll off characteristics of the highlight. The larger patterns which are further away from focus represent the ‘Bokeh’ characteristic. So Bokeh and focus roll off are linked issues. Notice how a medium sized aberration, in this case +/-0.035mm, has a large effect on through focus aesthetics. Hence is one reason why this author believes MTF to be a poor measure of image quality.
The basic cause is Schlieren photography. A small highlight acts as a point source. Ray tracing through a lens causes light to either concentrate or be dispersed at the sensor image plane. The effect reverses on the opposite side of focus; which is a key indicator of a phase effect in the lens. The highlight intensity is proportional to ray density, which is proportional to wave-front curvature. I have seen curvatures lower than 0.1 microns have a strong effect in test images.
In the example shown below, taken from the web under open permissions from Ashley Pomeroy, the lights simulate Schlieren photography with hard edges due to lens aberrations.
By normal optical theory, the hard edged ring is convolved with image information to create artefacts in the background. Hence, every time there is a transition from light to dark in the background, the feature has a hard edge. For example, the stair cases. Sometimes cinematographers like this and sometimes they do not.
But quality lens sets need the same Bokeh and focus roll off for every lens in the set so that images taken from different angles by different lenses can be matched. Therefore the higher order aberrations must match across a range of focal lengths.
This requires a tighter aberration compensation than is offered only by a good MTF. Notice again from the sample image how the higher order aberrations at the edge of the field of view may be important.
From an optical engineering point of view, Bokeh acts like a built in interferometer in every lens. To use it, it is necessary to image an out of focus point highlight, like an LED Christmas tree light. The result is a very sensitive ‘interferogram’ that measures less than 0.5 microns wave-front error. Though to get a quantitative measure requires a lot of interpretation, as the Bokeh is measuring localised wave-front curvature. One key indicator that one is looking at aberrations rather than some other effects occurs because the phase errors in the lens reverse the distribution in intensity across the Bokeh on the opposite side of focus.
A lot of lenses have zonal aberrations that induce noticeable hard out of focus edges in the distant background. However, the aberrations are also changing the ‘smoothness’ of the gradual softening of an image as the lens goes out of focus. The best lenses have smooth focus roll off with small amounts of zonal aberrations.
And the zonal aberrations are matched to give a consistent focus roll off over the whole range of lenses, which demands a higher quality standard than just achieving a reasonable MTF.
The third issue I am discussing is colour matching.
Aberrations affect how colours are reproduced. Not just in obvious ways. Colour balancing requires the red, green and blue signals from the sensor to be matched in a consistent way. For example, if the green signal is softened through aberrations in comparison to the red and blue signals, then the lens will present a false colour balance to the sensor. The differences should be matched across a complete set of prime lenses so that the images look the same.
Different lens types have different reputations. For example blog comments on Zeiss lenses include the comments ‘flatter’and ‘higher white contrast’.
Cooke lenses are called ‘rounder’ with a greater colour depth.
An extreme test of colour matching is to image reflections from scrunched up foil, as with this example from the SALT III lens tests. These images are from the two most consistent lens sets in the tests. Both look similar at T/2.8, though with a slight colour difference.
However, at T/1.4 there is a substantial difference between the lenses. But there is an even greater difference between images produced by the same lens at different apertures. In both cases, one would assume that images captured by the same lens at different apertures would require post processing to colour match. The variations between aperture settings indicate strongly that the difference is due to lens aberrations.
The images are courtesy of Matt Hayslett et al.
The effects of colour matching are most obvious when the images are shown side by side.
Let’s return to the Big OLE lens test party
(please double click on the separate video clip ‘SidebySide-windows.wmv’)
If one views the blog, then the comments by the lens testers show that they recognised a difference between these images. This is despite the scenes being shot at T/4. Assuming the differences are due to aberrations, then the aberrations are small, but still produce a visible effect.
This implies that intercut lenses must be aberration balanced to very fine consistency, and that MTF is not accurate enough to define the quality standard.
The owners of the video clip are identified from the credits.
Two still images where captured for further analysis. A line of pixel values was taken across similar parts of each image, shown by the yellow lines. The lenses were the Cooke 35mm and the Zeiss Superspeed 35mm. One would expect that the effects of aberrations in the lenses are well softened and suppressed by the T/4 aperture and the low video quality reproduced by the web.
Various factors could also affect RGB signal levels. For example, differences in brightness and contrast caused by differences in the angle of the face to the lighting and differences in the actor’s position relative to his ‘marks’. And one would expect spectral transmission differences between the lenses to change the relative levels of the RGB channels. One way of attempting to find a pattern, despite these variations, is to look at the differences between adjacent pixel values across the lines, as shown.
There are differences between the lenses, but no obvious relationship to lens aberrations. Isolating patterns that are only due to colour matching aberrations requires further processing.
I have therefore used simple mathematics in an analogous way to CIE chromaticity coordinates. At each pixel, an approximate equivalent to CIE colour coordinates is calculated, by taking one signal and dividing it by the sum of all three signals. The pixel to pixel differences of these coordinates is calculated.
By plotting out the blue and red constructs relative to the green construct, one can see a more pronounced relationship of the colour coordinate variations in the Cooke lens. One could intuitively relate this to the aberrations in the lens designs.
Because the analysis indicates that there is a better correlation between red and green and blue and green pixel values when imaged with the S4 lenses, and because the lens testers are commenting on the better skin rendition of the Cooke lenses, I hypothesise that a better rendition of skin tones is caused by the red and blue contrast changes more faithfully following the green contrast changes. And that this is caused by the matching of red green and deep blue aberrations in the lens design.
In theoretical analysis, I am using public domain information of the RGB absorption spectra of typical Kodak film. The absorption peaks correspond approximately to Ceg wavelengths, which are:
e: 546nm g: 436nm C: 656nm
The 436nm g wavelength represents a deep blue wavelength.
To follow this argument, I have set up two lens models. One is a lens taken from a Zeiss patent to represent a 50mm Masterprime lens. The second is a guess at a Ceg corrected lens (on axis only) with better deep blue correction.
A white bar pattern is computed from red, green and blue line spread functions, as shown below. The Ceg lens has a higher contrast blue pattern and a closer match between the red and blue images. The Masterprime image has a higher contrast red image and a closer match between the green and red images.
The discrepancy between the colour described by the text and the colour of the line is just a test to see that the reader is still awake.
Plots from the bar patterns show that the Ceg lens has a greater correlation between the changes in G/(G+B+R) and B/(G+B+R), except for two outliers. The parallel lines in the Masterprime model are caused by aliasing in the model.
I propose that this hypothesis has some merit and requires further study to evaluate. However, it is reliant on the artistic preferences of cinematographers, and hence on subjective comments about image quality. Therefore, it is difficult to define a method of deriving objective measures.
To summarise focus marks, they are required to measure focus consistently within the tolerance imposed by the Depth of Field. This is usually judged by circle of confusion of 0.01-0.025mm on an S35 sensor.
There are variations between lenses and aperture settings. But the lenses in a set require a consistent ‘look’ or aesthetic of focus roll-off. And the through focus MTF peak should be consistent with aperture setting.
Lens designs are sensitive to zonal aberrations, which should be similar across all lenses in a set. This lack of differences differentiates lens aberrations in the mathematical sense and demands a higher standard of aberration compensation.
I suggest that a suitable metric for focus ‘quality’ has not been defined, particularly with regard to focus roll off.
To summarise ‘Bokeh’, the consistency of aesthetic between shots is very sensitive to small variations in wave-front curvature. And that aesthetic is important across most of the width of the field.
Therefore a higher standard of aberration compensation is required to match different lenses and aperture settings.
I suggest that a metric suitable to measure Bokeh quality does not exist.
To summarise colour matching, the SALT III tests show an inconsistency in colour rendition from wide open to 2 stops down in any one lens.
There are subtle but noticeable colour differences that exist between different lens types wide open and at 2 stops down. There are greater variations between the Masterprime and Summilux C lenses when stopping the same lens down from T/1.4 to T/2.8, than between different lenses at T/4. Therefore, colour aberrations have noticeable effect with aperture variation.
The effects of differences between aberration states demand a higher state of aberration compensation, particularly zonal chromatic aberrations.
I suggest that a suitable metric for colour matching does not exist.
Three areas have been described where quality cine lenses have to be designed and built to standards higher than just a good MTF
Users vary aperture and lenses in one photographic sequence and are sensitive to variations in ‘look’. Users see lens performance differently from optical engineers and interact with subjective and artistic views.
Insufficient metrics are available to enable the two groups to interact objectively.
I wish to credit and thank the people below for granting permission to use their work in this presentation, as well as the people credited on the Big OLE Lens Test Party clips.
Picture of a Monkey statue at the Natural History Museum: Ashley Pomeroy via Wikipedia, open permission granted
Big Ole lens test party courtesy of Ben Eckstein et al (Other contributors are acknowledged in the above videos)
SALT III lens test courtesy of Matt Hayslett et al