keeney@vixvax.mgi.com (Richard Keeney) (05/04/90)
In article <20674@versatc.versatec.COM> ritter@versatc.versatec.COM (Jack Ritter) writes: >I am considering buying a digital film >recorder. I've seen 1 for about $5K. >It is a desk top model, with 4K lines >of resolution. By digital I mean: it >inputs a targa file (or similar image file), >NOT analog video. It's the Matrix Pro Color. >(Matrix was recently bought by Agfa). >At 4K scan lines max, I can print an image sized >up to 3500 high by 4000 wide pixels. (35mm film). > >That's right: 3500 X 4000 X 24 bits!! > >(Of course I'll never be able to see the whole >image an any CRT, but's taht's OK). > >This seems like a real deal. Any advice? Mr. Ritter is obviously not very familiar with digital film recorders (also sometimes called digital image recorders). For his benefit and others, I can supply a brief introduction to these devices in general. The most common type of digital film recorders used in the computer graphics industry basically consist of a CRT, some optics, a film transport, and some control circuitry to run everything. The CRT is typically high resolution, monochrome, and flat faced. Colors are obtained on the film by imaging three exposures of the CRT through three colored filters - Red, Green, and Blue. The scan of the electron beam is usually much slower than a display CRT for two reasons: To get the necessary anmount of light to expose the film, and because getting the digital data into the machine and through the modulation hardware (typically a digital to analog converter) can only go so fast. Other technologies exist like laser drum scanners, but I am not as familiar with these. Perhaps someone else can fill in here? Several manufacturers retail a variety of models of CRT based digital image recorders ranging in price from about $5000 on the low end to over $250,000 on the high end. The variety of models and the large price range reflect the large variety of needs of the users of these devices. A brief list of the major manufacturers of CRT digital image recorders: AGFA - Matrix, Orangeburg, NY Celco, Mahwah, NJ Dicomed, Minneapolis, MN Lasergraphics, Irvine, CA Management Graphics, Minneapolis, MN Mirus, Santa Clara, CA Pansophic (who purchased Genigraphics), Liverpool, NY Presentation Technologies, Sunnyvale, CA Polaroid, Cambridge, MA (My apologizies to any manufacturers I may have overlooked in the above list). The $5000 units with 35mm, 36 exposure cassette load transports, and 2k or 4k maximum addressable resolution (more on that later) are appropriate for occasional personal or departmental slide production. They generally can image a 4096 by 2732 by 8 bits per pixel by 3 pass image in 10 to 20 minutes. Some units under $10,000 offer interchangeable film transports that allow a variety of film formats to be used. Caution is advised here when considering these units for 4x5 or 8x10 work - the image quality is a function of much more than just the film size. The entire system has to be designed around the quality level generally required for large format work. One exception to this is when simple color overhead transparancies are to be produced - here, image quailty is not as important since the projector is going to be pretty poor anyways. Most units in $15,000 to $20,000 and up range offer interchangeable film transports and can produce reasonable quailty on most film formats. The ease of changing format varies greatly, so take into consideration how often you will be changing formats when evaluating these units. These units are the low end of what can be considered the "professional" range. Some key features available in this price range are bulk loaded film, pin registration, and imaging times under 2 minutes for a 4k slide. Bulk loaded film is important as imaging rates go up. When imaging lots of slides, who wants to change film every 36 exposures? Also, certain forms of processors similar to those used to process movie film can directly accept bulk rolls of film making processing of the film much easier for the service or department using such a device to serve multiple customers. Bulk loading is generally required to image computer generated movies. Pin registration is required to maintain frame to frame registration of the image. In order to be useful, the slides must be mounted in pin registered slide mounts. Accuracy of this level is required for professionally produced multi projector shows and is a basic necessity when imaging movies. At about $40,000 to $50,000 you can get an image recorder that will put 8k pixels per scanline on the film. Also at this point, imaging times should be under 60 seconds for a 4k slide. At $60,000 to $80,000 the image quality begins to approach that necessary to do a good job on 4x5 and 8x10 film. Also in this price range are units that can image a slide in under 30 seconds. Around $100,000 the 16k image recorders are available. This brings up an important point on terminology. Typically, CRT image recorders are specified as having a resolution of 2k, 4k, 8k, or 16k. This is in reference to the number of pixels that can be imaged on a given scanline of an image (powers of 2 are popular because memories for the line buffers work out easiest that way). Sometimes, people misuse the terminology and state that a 4k image recorder has "4k lines of resolution" or "4000 lines of resolution". There are two problems with this. First, "lines of resolution" usually means distinct lines - black lines on a white field for example. To do this on a device with discrete pixels requires that some pixels be white and some be black. Thus on a 4k image recorder only 2048 lines can even be attempted to be imaged since every other pixel would be white and the remainder black. The second common trap is to assume that because the pixels can be clocked out at a rate of 4096 per scanline, or the lines scanned on the CRT at the appropriate density to produce 4096 scanlines that the entire system is capable of resolving resolutions of 2048 lines. This is often not the case and is frequently intentionally not the case. Since a CRT produces a spot or scanline with soft edges (the energy distribution is commonly approximated as being Guassian), it is necessary to pack the pixels and scanlines together such that the spot overlaps the adjacent pixels in order to accomplish filling in of solid areas without having the individual scanlines be visible. In addition, since the CRT spot is soft edged, it is difficult to say how big it is without further qualification. Two common points to measure the width of the spot are at the point where it falls to 50% of its peak or at 1/e of its peak amplitude. Since these two are quite different, it is important to specify which one is being used. Another common misconception is that resolution = spot size / image width. A better statement would be that resolution is roughly proportional to spot size / image width. The exact limiting resolution depends on the exact intensity profile of the spot as well as non CRT factors like the optics, the film, and the electronics driving the CRT (rise and fall time of the video circuit for on-off transitions greatly affects the horizontal resolution for example). One important factor in imaging times that people often overlook is how fast the data can be transmitted to the image recorder from the host machine. Typically, the data file describing the image resides on the disk of the host machine. With a 4096 x 2732 x 24 bit image, this file is 32 Mbytes long. Even with a fairly fast net transfer rate of 500 kbytes/second, it takes at least 64 seconds just to copy the file from the disk to the image recorder. At higher resolutions, or with more bits per pixel, the transfer times go up dramatically. Some image recorders allow compressed forms of the image data to be transferred to help alleviate this problem (but now CPU time must be expended to compress the image). At 8192 x 5464 x 36 bits, the image file would be 268 Mbytes (assuming the 12 bits per pixel occupied 2 bytes and were not packed), and at 16384 x 10928 x 36 bits, the image file would be about 1 Gigabyte long. When deciding on purchasing a digital image recorder, it is best to not rely too heavily on manufacturer published specifications since they are often vague and intentionally misleading and instead look at some actaul film produced by the model(s) you are interested in. Since film is relatively cheap, most manufacturers will readily give away samples to qualified prospective customers (those with money and who are actually going to buy something). If possible, you should get your hands on a demo or loaner unit and do some testing yourself based on your specific application. Dave Martindale of IMAX systems published an excellent article here a few days ago outlining several simple tests to perform on a film recorder under consideration. I will not reinvent the wheel and will simple include his article here: In article <1990May1.192338.7449@imax.com> dave@imax.com (Dave Martindale) writes: >[...] >Draw a pattern of vertical lines, with all even-numbered pixels >black and all odd-numbered ones white. See if you can actually >see the black/white alternation on the finished film. (I'll bet >you can't!) Do the same thing for horizontal lines. > >Then try for 2K effective resolution: two pixels black, two pixels >white, two pixels black, etc. See if you can see *that* on the >film. Again, do it horizontally and vertically. Keep increasing >the size of the bars in the pattern until you can see it on film - >this will give you some idea of the resolution limit of the system. > >On a pattern where the black/white bars are clearly visible, >compare them at the centre, edges, and corners of the image. >This will tell you how much things degrade in the corners. > >Plot a full frame of several different grey levels. Look at the >resulting film for any detail that should *not* be there - vertical >lines, horizontal lines, ghost images, etc. > >Plot full frames of the primary colours, to see how pure they are. >Plot full-intensity white, to see how much exposure it can get onto film. > >Plot a full-intensity coloured square in the centre of a grey frame, and >see how much of the colour "spills" outside the square into the grey. > >Measure how long it takes to actually record 1 frame at 4K. > > [...] These are all very reasonable tests for evaluating a digital CRT image recorder. I would like to point out some additional areas of concern when evaluating a CRT based digital image recorder: Another important thing to evaluate is the image geometry. That is to say, is the image the correct size, in the correct postion, level, and not distorted (edges curving in or out)? On CRT based image recorders like the AGFA Matrix ProColor and many others, getting the image geometry right is no easy task for the manufacturer. The basic problem is inherent in the process of making a CRT out of glass - you just cannot hold all that high of tolerances in the positioning of the gun in relationship to the screen and the yoke when dealing with molten glass. Most CRT image recorders have numerous adjustments for getting the geometry right. In some, this takes the form of trim pots, and in others, it is accomplished digitally. To test image geometry, image a pattern with fine (1 pixel thick) lines through the center of the image vertically, horizontally, and at 45 degree angles. Also include a border around the edge a few dozen pixels in from the outermost pixels (since the outermost pixels may be clipped by part of the film transport). Check that these lines are straight and that the border is the size you think it should be. A microscope with a micrometer X-Y stage would be helpful in measuring the film. Another issue is the quality of the color modulation of the device. Typically, digital CRT image recorders control the exposure of pixels on the CRT in one of two ways: By either modulating the voltage applied to a grid of the CRT using a DAC (digital to analog converter) thus modulating the intensity of light emitted by the CRT, or by varying the amount of time that each pixel is turned on with the CRT emitting a constant amount of light. In either case, the mapping function between the raw numbers used to control the exposure and the resulting color of the pixel on the film is unlikly to be linear, or even all that simple (the film itself helps to make this even more difficult to characterize exactly). In order to obtain the desired mapping betweenthe digital codes fed into the recorder and the actual color produced on the film, a remapping or look-up-table is used inside of the image recorder. This table takes the user's color values (typically 8 bits per primary per pixel) and maps it to the numerical code required to control the modulation system of the image recorder. The quality of this remapping along with the underlying precision of the modulation system of the image recorder combine to determine the overall accuaracy and precision of color reproduction of the image recorder as a whole. I recommend two test images to help evaluate this part of an image recorder's performance. The first would be comprised of a series of 10 to 16 grey patches varying in intensity from black to white. Check that all of the patches are a neutral grey and are not slightly tinted. Usually, if a tri-color system can get the neutrals right, the other colors will fall in line as well. A densitometer like those used in photo processing labs to check the colors of the processing machine can be used to measure the exact colors obtained if the patches are large enough. Remember that processing and the exact emulsion batch can have a moderate effect on the color of the resulting film. The second color reproduction test that I would recommend would be to create an image comprised of a complete gradation from black to white. On an image recorder with 8 bits per primary, this would be accomplished by dividing the image into 256 equally sized strips and making the first one 0:0:0, the second one 1:1:1, ... 254:254:254:254, and finally 255:255:255. The steps on the resulting film should almost not be visible (8 bits per primary is not quite enough to make the steps completely invisible - that is why some image recorders use as many as 12 bits per primary). There should be no hue differences between adjacent steps. An image recorder that has problems on this test will cause scanned or synthetic 3d objects to have contours, but may be fine for 2d object based graphics (like white text on a solid blue background). The final thing that I would like to mention concerning evalutation of CRT based digital image recorders is not so much concerned with the quality of any one image, but with how repeatable that quality is. CRT's and their related electronics tend to drift with temperature, and also to age. If not corrected for in the design of the image recorder, the long term repeatability of the image recorder will suffer. This is important when an entire show (or movie scene) is shot and a few weeks later, one slide (or an adjacent scene) is reshot and the colors of the newly shot film are a little "off" from the colors of the older film. Long term stability is not an easy thing to test quickly. One way to get an idea of how well an image recorder will perform in this regard is to look at film shot on several different units of the same model. If the colors are well matched, chances are, the image recorders in question cannot be drifting too much, or it would be difficult to get them to produce the same color. This test has the obvious weakness that the manufacturer could simply hand pick several units that happen to match. As an employee of one of the major manufacturers of CRT based digital image recorders, I have attempted to remain unbiased in the above discussions, but to justify the time I spent on this, I guess I will have to include the following paragraph: Management Graphics, Inc. retails a line of Digital Image Recorders called Solitaire. Currently four models are available. A brief summary of the models: Solitaire-4: Our lowest cost model. 4k slides in 120 sec. Solitaire-8: The original. Up to 8k. 4k in 45 sec. Solitaire-8xp: A new improved Solitaire-8. 4k in 29 sec., improved large format image quality. Solitaire-16: Up to 16k. Spot size 1/2 of other Solitaire's. Contact Bill Gillooly at +1-612-851-6124 for further details. -- Richard Keeney Internet: keeney@mgi.com Management Graphics, Inc. Phone: +1-612-854-1220 1401 East 79th Street, #6 Fax: +1-612-854-6913 Bloomington, MN, 55425 U.S.A.
ritter@versatc.versatec.COM (Jack Ritter) (05/04/90)
I wish to thank Richard Keeney for his recent,
very thorough discussion of digital image recorders.
This subject is becoming more & more popular as
film recorders go down in price. I have been collecting
articles in comp.graphics about film recorders for
about a year now. It's all in one file, including
Richard's. I will post it if enough people ask for it.
About my own quest: I'm now looking at 12-15K.
Everyone please send me 20 dollars.
--
Versatec, Inc.
Jack Ritter, M.S. 1-7
2710 Walsh Ave.
P.O. Box 58091
Santa Clara, CA 95052-8091
(408)982-4332, or (408)988-2800 X 5743
UUNET: {ames,apple,sun,pyramid}!versatc!ritter
--( / __ - .. (( /
/ / -- ) . \ \ // . (
/ ** ) // _*_ // * ..
) (( . \ / . * ) //