Session How Many Pixels In Lawrence of Arabia?
Presenter By Dr. William Glenn, Florida Atlantic University
John Galt, Panavision
James Pearman, Panavision

ABSTRACT

As archivists and restoration specialists are now using digital imaging systems to provide a digital record, and often use digital imaging tools to "restore" lost or degraded image content, it will become increasingly important to have the means to assess the data content of images prior to digitization.  This will be important both to ensure that the techniques employed do not reduce the information content of the original images and, equally important, do not over-sample unnecessarily.  The latter is critical because many images of historical or cultural importance may have no immediate or obvious commercial value and the cost of digital archiving and restoration will often be a determining factor in deciding whether or not to digitize and "restore" the image content.

Imaging specialists have many tools to examine the performance of lenses, film emulsions, cameras, etc., and to predict the cascaded performance of imaging systems. However, these techniques, useful as they are in predicting final image content before the fact, tell us nothing about the information content of the existing image.

This paper will first examine the latest techniques used in image system analysis using, as an example, the original camera and lens system employed in the photography of "Lawrence Of Arabia".  This will be correlated with an experimental technique used to determine the information content of existing images.

_______________________________

PRESENTATION

As archivists and restoration specialists are now using digital imaging systems to provide a digital record, and often use digital imaging tools to "restore" lost or degraded image content, it will become increasingly important to have the means to assess the data content of images prior to digitization. This will be important both to ensure that the techniques employed do not reduce the information content of the original images and, equally important, do not over-sample unnecessarily. The latter is critical because many images of historical or cultural importance may have no immediate or obvious commercial value and the cost of digital archiving and restoration will often be a determining factor in deciding whether or not to digitize and "restore" the image content.

In both the digitizing of film images and digital imaging in general there has been a tendency to describe resolution in pixels i.e. the film has been digitized at 2K, 4K, 6K etc, or it is a three, six or twelve megapixel camera. This focus on a single parameter of image performance may be considered to be a surrogate for image quality but it is a very limited one. Final image system performance of a film or digital image is a cascade of many imaging system parameters that can include pixel count, Modulation Transfer Function (MTF), dynamic range, signal to noise or granularity, color gamut and a variety of less obvious parameters such as lens spectral bandwidth and pixel fill factor. None of these parameters can individually be used to predict final image quality. Image parameters are often mutually exclusive, emphasize one and another will suffer. For instance it may not be immediately obvious that increasing color gamut can decrease resolution or an imager with more pixels can have less dynamic range. It is also important to realize that final image quality is a cascade of all the elements in an imaging system and that final image quality is always less than the performance of the lowest performing element.

This paper will explore the function of the various system components and their contribution to final image quality that equally applies to both film and digital imaging systems. Our intent is to demonstrate that an ideal motion picture imaging system, whether film or digital must achieve an appropriate balance of these often-contradictory parameters.

Modulation Transfer Function (MTF) is the most commonly used metric to describe both single element image performance, such as a lens and also cascaded final image performance that could include the principal photography lens, imager (film or CCD), intermediate optics such as printing lens, output film or digital projector imager and projection lens.

This graphic demonstrates the concept of Modulation Transfer Function. In this example a high contrast bar target of increasing spatial frequency, i.e. the bars are both smaller in horizontal dimension and distance apart is photographed. The ideal response of the imaging system would be 100% contrast for all frequencies or image detail. That is, black would be 0% contrast and white would be 100% contrast. However, the real response is far from ideal and it can be seen that with increasing frequency or smaller detail the contrast becomes less and less. By graphing percent contrast versus frequency a Modulation Transfer Function curve can be derived.

The special property of MTF measurements is that they can be cascaded or multiplied such that a complete picture of an imaging systems performance can be derived by cascading all of the individual system MTFs.

The resolution performance of imaging systems is often exaggerated by considering limiting resolution (or threshold of detection) as the sole description of image quality. Limiting resolution, usually described in line pairs per millimeter, relies on the assumption that an observer has unlimited time in which to determine whether there is a visual difference between two imaged areas of a high contrast target. The United States Air Force (1951) test target epitomizes this approach to measuring image quality.

In an ideal observation environment, a visually normal observer can detect very small contrast differences. Thresholds of Detection Measurements are most significant in applications such as aerial reconnaissance photography. They have limited value in motion picture imaging applications.

Motion picture photography usually involves photography of three-dimensional objects in a three dimensional space. Photographers and cinematographers use Depth of Field as an established metric that dictates the acceptable limit of image performance. Since most photography involves three-dimensional objects, and a lens can only render a single plane in perfect focus, the concept of depth of field is that objects sufficiently close to the plane of focus will appear sufficiently sharp to appear to be in focus. Photographers rely on depth of field calculations and charts to determine if objects will be appear acceptably sharp in a final image.


This graph represents the MTF performance of two lenses one of modern design and the other a 40-year-old design. Both lenses were carefully focused at eight feet and MTF measurements were made at various distances in front of and behind the plane of focus. From depth of field tables, and assuming a circle of confusion of one thousandth of an inch or 25 microns, it can be seen that at the depth of field limit the MTF of the lenses has dropped to between 35 and 50%.

From the measured depth of field performance of various lenses it is obvious that an optical performance of less than 50% of maximum MTF must be considered to be the limit for motion picture imaging. It is also important to understand that perceived resolution is also a function of magnification. A higher resolution imaging system requires both a reduced depth of field (i.e. smaller circle of confusion) and a reduced audience distance to screen size ratio. Although beyond the scope of this paper, a depth of field metric based on the specific MTF performance of a lens would be somewhat different from that implied in the classical depth of field formula and tables.

MTF measurements can also be misleading. MTF measurements are often made in the center of the field at a defined aperture. This will tell nothing about the off axis performance of a lens. In addition, MTF measurements are often single wavelength measurements such as the “e” ray that will usually show the highest MTF for a lens designed for visible light photography. A lens with better chromatic correction over the whole normal visible spectrum could be disadvantaged in a side-by-side comparison under this restricted condition.

Having given all these caveats MTF measurements are still the most important measure of individual imaging elements and system performance. This is because MTF measurements can be cascaded to provide a complete system MTF. It is the cascade of all the system elements that we ultimately must be concerned with. It is also important to recognize that system performance will always be worse than the lowest performance element in the chain.

To demonstrate that limiting resolution is not a valid metric for a motion imaging system, we created the following demonstration.

The following series of images illustrate image performance at approximately 25 cycles per millimeter of 100%, 50% and 5% depth of modulation. The 50% modulation image is significant because this usually is considered acceptable image performance at the limit of depth of field.

As previously stated the Modulation transfer function, or MTF, from either an MTF optical bench system, or an imaged sinusoidal pattern provides a more comprehensive measurement of imaging performance, and one which better agrees with subjective evaluations of continuous tone images. Imaging a sinusoidal pattern can be used to measure the cascade of a lens and image forming system. The image forming system can be either film or a CCD array. The sinusoidal pattern shown here is a commercial product available from Sine Patterns LLC in Pittsford New York.

The resulting waveform is somewhat difficult to analyze due to amplitude ambiguities. Ideally, we would like to be able measure the average amplitude of each group of sine frequencies.

The chart shown here is a recent addition to our MTF measurement arsenal. Primarily designed to automatically generate MTF curves in digital imaging systems it can be equally applied to Digital Intermediate Processes. The chart consists of a series of discreet sinusoidal frequency patches and other calibration patches. In use the chart is imaged so as to occupy one third of the horizontal field of view. The resulting digital image, or digitized image if we are measuring a film negative through a digital intermediate process, is processed with software designed to automatically derive a RMS (Root Mean Square) value for each group of sinusoidal frequencies and then plot the resulting MTF curve.

The following series of cascaded MTF curves represent the best performance attainable with modern film, optics and digital imaging systems.

The first graph shows the best performance we can expect when we combine a high performance lens with a modern high-resolution medium speed (EI 250) negative.

The second graph shows the addition of a printing lens and already we can see that the MTF at 80 cycles/mm or 4K pixels has dropped to 5%.

The final graph shows the addition of an intermediate film emulsion. This would represent a negative scanned at 4K with a laser printer also output at 4K generating an inter-positive or printing negative.

It must be emphasized that these are idealized results that will not be met in any existing practical system.

These techniques, useful as they are in predicting final image content before the fact, tell us nothing about the information content of the existing image.

One can measure the modulation transfer function of film by scanning it with a laser scanner and looking at the signal with a spectrum analyzer. This is a secondary measurement since it depends on the accuracy of calibration of the scanner and analyzer. It also does not give the angular distribution of the spatial frequency information. This distribution can be very important if the film is digitized or displayed with a digital projector.

In the experiment described here the modulation distribution is measured directly on the film using optical processing. This gives both the modulation amplitude and direction of orientation of the spatial frequencies. The optical setup is shown below.

A laser is used to illuminate the image on the film. A beam expander is used to expand the laser beam to cover the area of interest on the film. A lens near the film is used to focus the light from the laser onto a CCD sensor. The light distribution on the CCD is the two dimensional Fourier transform of the information on the film. With no film in the film gate the focal spot on the CCD represents the DC component of spatial frequencies in the image. Any detail will diffract light by an angle that is proportional to spatial frequency.

Results

With film in the gate with a mid gray density and no image there is a circular haze around the central focal spot. This represents the spatial distribution of film grain noise. With an image on the film there is a distribution of light diffracted by the information in the image. Several things are noticeable in this distribution. When the amplitude of the image information gets below the haze (grain noise) the only useful information on the film is below that spatial frequency. The system was calibrated by recording sinusoidal patterns on high resolution film with a laser scanner. Even though the frequency driving the laser scanner was very constant, the spatial frequency on the film varied slightly from side to side due to a slight scan nonlinearity of the laser scanner. This is a good way to check geometrical linearity of scanning.

For image information it was also noted that the angular distribution of the information was not symmetrical. There was much more information in the vertical and horizontal direction than in oblique directions. In natural scenes this is due to the direction of gravity. Long thin objects have two stable orientations-standing or hanging straight up or lying flat on the ground. Tree trunks are vertical and the horizon is horizontal. Man-made objects have an even more preferred orientation. The edges of windows, doors and bricks are either vertical or horizontal. All of the elements in a seven segment LCD are either vertical or horizontal. In addition, the human visual system has better acuity in the vertical and horizontal directions. This is referred to as the “oblique effect” in vision research. For this reason printers have used halftones on a diagonal in printing since this produces better vertical and horizontal resolution than dots in rows and columns. This also is advantageous in digital sampling patterns of images. Diagonal sampling is preferable over cardinal sampling for the same reasons.

The film “Lawrence of Arabia” was chosen for this test since extreme care was taken in the design of the camera lens and film transport to produce images with outstanding quality. The film had a reputation for outstanding images in the theatre. Original negative was analyzed to see at what resolution the image information became buried in the film grain noise.

The graph shown is the MTF performance of one of the lenses used in the photography of “Lawrence Of Arabia”. Despite the fact that this lens is over 40 years old its imaging performance rivals that of the most modern lens designs today. This is not because of poor performance of modern optics but because the large area of the 65mm negative means that the effective image performance is more than twice that of the smaller 35mm negative.

Although the performance of the film emulsion used in “Lawrence”, (Eastman Color Negative Type 5250 with an exposure index of 50 in daylight and 32 in Tungsten illumination) is easily eclipsed by modern T grain emulsions, again we can see from the cascaded performance that the large negative still results in a final image performance that is not readily attainable with modern optics and 35mm format negative film.

So what is the answer to the question, how many pixels in “Lawrence Of Arabia”?

From the MTF characteristic of the negative it would appear that digitizing the negative at 6K pixels horizontally would capture all useful information. Considering the negative aspect ratio this would result in approximately 16 megapixels per color or 48 megapixels per frame. Running time is approximately 227 minutes. At 1440 frames per minute this would suggest about 16 million megapixels would be sufficient!

_______________________________________

SPEAKER BIO

John Galt

John Galt has been employed for the past five years at Panavision, Inc. He is currently the Senior Vice President of Advanced Digital Imaging at Panavision’s corporate office. His responsibilities at Panavision include the development of digital imaging technologies in support of Panavision's core motion picture and television production business. Most recently Panavision's Advanced Digital Imaging Group has introduced the first 24P Digital Electronic Cinematography System.

John Galt was previously employed as Vice President, High Definition Technology Development for Sony Pictures High Definition Center. His main responsibilities were the integration of electronic and film imaging systems. This included film preservation, high definition film transfer systems and electronic cinema. John was project leader of the group, which designed and built the first High Definition Telecine in North America.

Prior to Joining Sony in 1988 Galt was president of Northernlight & Picture Corporation in Toronto, Canada. Northernlight & Picture was the co-producer along with the Canadian Broadcasting Corporation of "Chasing Rainbows" a fourteen-hour drama series, the first to be produced and photographed using high definition video technology. John Galt was also Director of Photography on this project.

He holds several U.S., British, and Japanese patents in film and electronic imaging related areas.


William E. Glenn

Dr. Glenn received his PhD from the University of California at Berkeley. He is presently Director of the Imaging Technology Center at Florida Atlantic University. Previously, his early work on digital video compression of HDTV at New York Institute of Technology was credited for starting the conversion to digital television by Joel Brinkley’s book “Defining Vision”. In his previous position as Vice President and Director of Research for CBS Laboratories his invention of the digital noise reducer won an Emmy award. Before that at the GE Research Laboratory he invented and developed the “Talaria” light valve color television projector. This was a successful product for 26 years. In his further work there in 1962 he used a 1050 line progressively scanned camera to record and display HDTV using a recording and display system that he invented. He has 105 publications in the field of electronic imaging and holds 123 issued US patents. Approximately half of these patents have become commercial products.