1. The quality of the media. For magnetic media: How close the particles are, and how thick the coating is. The more particles per area, the higher the resolution and therefore the more data that can be stored. As the coating thickness increases, so does the amount of data that can be stored before bleeding (or blurring) occurs. Lower quality media also shows wear faster ... as the magnetic material wears off, the data "signal" becomes weaker (the data resolution decreases and data loss increases).

2. The size of the media. Obviously, you can stuff more video information on a 3/4-inch or 1-inch tape than you can on 1/2-inch VHS.

3. The speed of the media. The faster the media is moving, the higher the bandwidth for a given period of time. Ie, 15ips on 1/4-inch audio tape has more data than 7ips.

4. The format of the data. How are brightness and color data stored? Do they coexist in the same media area or are they discrete? (See Evans' post above about color and luminance)

For DVD, picture data is stored in a pixel format. Each pixel has its own color and brightness value. DVD stores 720 pixels per line for 480 lines, resulting in phenomenal bandwidth requirements, even with compression.

In regard to using film to store digital data, it all depends on the grain resolution of the film stock and the methods of data recording and of data retrieval and processing. I wouldn't go with any type of color or greyscale, as your reader light might be dimmer than mine, causing a color value of FF (white) to read as 66 (medium grey) unless the reader is recalibrated to the light source with every use.

Then there's the issue of bandwidth. I seem to recall that the Wizard of Oz was reworked at a resolution of 4K x 3K. Using only binary data (black/white or off/on), that's only 12 megabits (or about 1.4 megabytes - think 3.5-inch floppy disk) per frame on an academy aperture. And that's with no redundancy or checksum for error correction.

In my opinion, that electron beam thing sounds too expensive for anything other than government use.

One can visually measure the horizontal resolution of a playback system by looking at the wedges of a SMPTE resolution chart, available on VHS (SMPTE), U-matic (SMPTE), LD (Video Essentials) and DVD (Avia, Video Essentials).

Typical luminance horizontal resolution numbers:
VHS about 240 lines.
Off-the-air NTSC about 330 (approx 80 lines per MHz).
Super VHS and LD about 400.

Chroma resolution will be about half that of the luminance, designed that way for NTSC/PAL or 4:2:2 since the human eye sees fine detail only in B&W (rods having a much smaller cross-section than cones).
Think of a video image as being like a comic book image--the fine detail is done with a pen, the colors are applied with a brush.
Home tape formats will be somewhat worse since they are all color-under systems.

Oops, once again I have talked too much.

While I do understand the concept of bandwith limitations, I don't quite see how it works with analog video -- it's not a continuous stream, it's divided into scan lines. And I assume there has to be some kind of pulse that signals when one line ends and the next begins. If so, does that mean VHS actually has fewer of these pulses per field/frame than full NTSC resolution? If so, how does that get converted to the full 525 lines for display? Or does it? Can TVs simply spread their scanlines out further and have fewer of them on screen (making them, in effect, like multi-sync computer monitors)?

Sorry, I'm really displaying my ignorance here. These must seem like stupid questions to someone who really knows about video. The explanations regarding picket fences and ladder rungs are interesting, but they don't really address what I'm having trouble with. I guess I need to just go to the library and see what I can find out. Thanks to everyone who replied, though.

Oh yeah, and I guess this thread which was supposed to originally be about digital formats has now been hijacked into being about analog video...

VHS, Laserdisc, S-VHS, and DVD all have 480 visible scan lines of information, and all output NTSC signals with 525 scan lines for an NTSC signal (with the same number of sync pulses). DVD actually encodes 720 pixels per scan line for the visible portion. The analog formats encode scan lines, but each scan line is a continuous analog signal and is not discrete (meaning it's not made up of pixels). A VHS picture is made of the same number of scan lines as the output of a TV broadcast, a Laserdisc player, an S-VHS player, etc, but the information that can be recorded on each scan line is less than all the others. It's not that VHS has fewer scan lines, it's just that each scan line can hold less information, resulting in a less detailed picture in the horizontal direction. The resolution in the vertical direction is the same as any other standard NTSC format. It's the resolution in the horizontal direction that suffers. VHS pictures are not made up of "pixels", and neither are S-VHS, Laserdisc, or any other analog video source.

------------------
Evans A Criswell
Huntsville-Decatur Movie Theatre Info Site

Okay, I said I was displaying my ignorance, but I'm not that ignorant!

I understand the difference between digital and analog video. I know that of NTSC's 525 lines, only 480 are active, and that DVD is 720x480 discrete pixels.

I also understand that while analog video formats are divided into discrete rows (i.e., scanlines), they are not divided into discrete columns. There are no pixels, each scanline is just one continuous (albeit bandwidth-limited) stream. One can assign a pseudo-value as to how many columns the lines contain (and say that there are "about 300" or however many), but it's just an approximation, and not an actual, discrete number. I get that much. I'm with you that far.

What I didn't understand is that some explanations (such as the web pages I linked to earlier) really did seem to imply that there are fewer rows (i.e., scanlines) in some analog video formats than others. When they talked about "horizontal resolution", I misunderstood and thought that meant there were fewer scanlines. I was wondering how that could possibly be, since the 525/480 number sounds as though it is fixed.

However, after reading Paul Mayer's post again (thanks, Paul!), I realizing I was just misreading and getting confused about "horizontal" and "vertical" resolution. I suspected that might be the case, and it makes a lot more sense to me now. Sorry about all the needless questions.

The trouble with WWW sites is you never know if the creator of the WWW site has sufficient technical knowledge. Someone can make a neat-looking WWW site and draw nice diagrams saying that the resolutions of different formats is such and such, even if the information is wrong. When you see such information, it may be right or it may be wrong. I'd trust information from a reference book more than from a random WWW site. I've seen incorrect information on WWW sites made by reputable companies.

Another problem is that some digital motion picture formats are lower resolution than NTSC. I have seen some TV commercials where it was obvious that the commercial was rendered for a fewer number of scan lines than standard NTSC, and some sort of scaling algorithm was being used. I see this fairly often, and wonder why they do it. Such a format may have approximately the same resolution as VHS (or more resolution that VHS). If VHS is roughly 240 by 480 (480 scan lines), then a digital format with a resolution of, say, 512 by 384 (284 scan lines) has more resolution, in fact, 71 percent more, but due to the processing required to change the 384 scan lines to 480 scan lines to make a standard NTSC signal, and the fact that broadcast TV is going to limit the bandwith of the signal to around 330 "lines of resolution" per scan line, a commercial rendered that way may look inferior to a VHS tape due to the artifacts created by the scaling from 384 scan lines to 480.

The interlaced nature of video makes things more complicated. You can think of video as consisting of 60 frames per second (except call them fields) at half the vertical resolution, since only half the scan lines get displayed on each screen refresh on a standard NTSC set. If material is recorded for TV with a standard TV camera, then there will be motion in every field. That is, the two fields that comprise a frame will be different. You probably don't want to stick those two fields together and encode them as a single picture with 480 scan lines in that case. In this case, each field can be encoded in sequence, each requiring a vertical resolution of only 240. Interlacing in TV cameras and monitors is a quite complex issue, complicated even more by digital video cameras and issues in digital encoding, as well as conversion of an interlaced video signal to a progressive one. I don't completely understand all of the issues involved myself, but I tend to look at each format in terms of how much resolution could be present in ideal cases. For example, an interlaced signal from a VHS tape could have 480 true scan lines of information since the 3:2 pulldown method of recording the movie would allow the fields to be weaved together to produce progressive images, since there would be no motion between fields that were scanned from the same film frame. You could not do the same with a VHS tape of a football game shot with a TV camera, where vertical resolution has been traded for temporal resolution. Interlaced video creates interesting issues concerning vertical resolution and temporal resolution.

However, even if a digital format has fewer scan lines than NTSC and can take advantage of the interlaced nature of an NTSC signal and "cheat" and get by with encoding a lesser amount of vertical information (scan lines) than 480, that looks just as good on a standard TV, I still argue that the signal is inferior to a signal with 480 scan lines encoded per frame, since such a signal, if from a film source, can be line doubled to produce a progressive signal, while the format with fewer scan lines encoded would require interpolation to display 480 lines, which introduces artifacts.

An interesting article is Interlace or Progressive Scanning by John Watkinson, Watkinson International Communications, available in Pixels, Pictures and Perception: The Differences and Similarities Between Computer Imagery, Film and Video from SMPTE.

Personally, I'd like to know more about the digital encoding of interlaced NTSC myself, and how systems that use fewer than 480 scan lines of information work.

I'm looking at the WWW site http://www.elitevideo.com/new2.htm and see one error already. .7 x 483 is not 240. Computer and video resolution is measured in terms of how many pixels (how many changed from white to black) can be stored rather than how many pairs of changes from white to black can be stored. Film folks say that a resolution of 480 is the ability to store 480 pairs of white and black lines, while video and computer folks would call this a resolution of 960, since that's how many pixels would need to be there. The Kell factor comes from aperture effects inherent in tube cameras and CRT devices. The Kell factor varies depending on the apertures of the systems in use and may not always be .7 . This Kell factor does not mean that the resolution of a given video format is less. It simply means that it may be less when displayed on a CRT, or may have been limited if a tube camera was used.
------------------
Evans A Criswell
Huntsville-Decatur Movie Theatre Info Site

It is interesting that consumer electronics have surpassed a national TV network.

Joe, in my opinion, home electronics surpassed broadcast TV (I'm talking standard analog) when Laserdisc came out.

About this interesting Kell factor:

I got my copy of Fundamentals of Television Engineering by Glenn M. Glasford, 1955, and this Kell factor is discussed on pages 26 and 27. There is a discussion of the position of the scan lines in relation to vertical detail, and the Kell factor being determined by either "subjective methods, or a more complex theoretical analysis". It mentions two different subjective techniques from which the Kell factor was determined to be 0.64 in one case (using a test pattern with a tapered wedge of almost horizontal converging lines) and 0.71 in another (based on reproduced with of a single line when crossed by the scanning lines). Using different test patterns leads to different Kell factors.

As I said before, the characteristics of the display device itself and even the camera used to record the images will also affect the Kell factor (and this is a separate ddiscussion from the one in the above paragraph).

But, when we speak of resolution, what the hell are we talking about? Different people mean different things when they speak of resolution, and I'm seeing that when I research this topic. The discussion from the Fundamentals of Television Engineering book, when discussing the Kell factor, is speaking of resolution in terms of how many pairs of alternating bright and dark lines can be displayed. Specifically, when speaking in terms of scan lines and the Kell factor, the question is, "With a certain number of horizontal scan lines, how many pairs of regularly spaced alternating light and dark horizontal lines can be discerned on the display device?" As a computer scientist, my question is "How many 'pixels' of information could I store on the picture in a column from top to bottom?" These are two different measures of resolution and in discussing either TV displays or the resolution of an image file, can produce nonintuitive differences in the answer depending on what is meant by resolution.

I'll tell you my opinion about this "Kell factor". It seems to be a subjective measure of how much interference from beat patterns is acceptable or not acceptable. Let's look at this "reduction of resolution by the Kell factor" in reverse. Let's say I have 32 pairs of alternating solid white (value 255) and solid black (value 0) horizontal lines. Obviously, with 64 scan lines, I can display these perfectly (assuming the display device is capable of displaying all the scan lines without the scan lines overlapping, which is an entirely different issue, that I'll get to later -- called aperture effects). Suppose I have to use 70 scan lines instead of 64 due to the display device requiring that frequency. A beat pattern will be created that will interefere with the information due to the resolution change (signal resampling). Although my display device can now handle 35 pairs of alternating white and black lines, it cannot handle 32 well at all due to the beat pattern caused by the resampling. So now the question is, how many scan lines do I have to use to create a display of the 32 alternating lines. I'm going to let each of you find that out on your own. Go here . Take the test and report your Kell value here. This will be interesting.

To say that a Kell Factor is 0.64 means that you have to use 100 scan lines to display the 32 alternating pairs of white and black lines and be able to discern them. The ability to discern is a bit subjective, and the characteristics of each of your monitors may be different, so I expect your answers to be different.

Now, the other thing that causes loss of resolution on a display device is aperture effects. An electron beam scanning across a picture tube is not a point, but rather a spot that has size and shape, and this size and shape affect the frequency response of the display device. Extraneous patterns are created if the scanning spot is so small that there is black space between the scan lines, but a scanning spot too large causes overlap in the scan lines, which results in a blurring of the image. This blurring would reduce the number of alternating white and black lines discernable.

So, if it is determined that on a given TV display, only 320 equally spaced horizontal lines are discernable, then it could be due to the Kell factor, aperture effects, or both.

However, I still stand by the claim that 480 scan lines of information must be encoded in the source material to get the maximum resolution out of the display device. If, due to Kell effects, 320 scan lines are discernable on a given display, and source material, as a result, is encoded with only 320 scan lines, thenthere first must be a conversion (resampling) from 320 to 480, which introduces a beat pattern in the signal and degrades it from its original form, before Kell and aperture effects take further toll on the signal when being displayed. You can't say that just because of Kell or aperture effects, that the resolution of, say, VHS, Laserdisc, or DVD is less than 480 scan lines.

The unfortunate result of all of this is that there could be 3 differenct resolution figures reported for the same device:

N (number of raw pixels)
N/2 (pairs of white/black lines displayable by alternating pixels)
K * N/2 (pairs of white/black lines displayable taking Kell factor into account)

Also see the thread entitled "Digital projection challenges - resampling and rotation" in the Film Handler's Forum from March 23, 2001.

Thinking of objectionable patterns in everyday television, even at low resolutions, these seem to be surprisingly rare, largely due to educated videographers avoiding potential problems. The Hitchcock Cary Grant film about the cat burglar is one (the name escapes me at the moment). Grant wears a herringbone jacket in the film and there are stripes in the shirts that moire badly. Other than that notable exception, I am rarely bothered by beat freq. artifacting on television.

More specifically, the technology of using scan lines is inferior in resolution to film vertically, but has the potential to be better than film horizontally. This is a known trade-off. The solution to scan line problems lies in finding a suitable pointalist substitute that can send a similar amount of information in the same size data stream, and not require an unworkable yoke circuit. Random dots with a certain amount and percentage of overlap could provide a more flm-like experience, but I can't imagine the circuits involved.

quote:

---

An improvement that would be within current technology would be to scan horizontally on one interlace and then scan vertically on the other interlace. This would increase the problem on one axis but reduce it on the other, and result in a _perception_ of a much clearer picture in both dimensions. The problem with this improvement is finding a way to change millions of tv sets...

---

Wow. If the interlacing was done this way, it would surely make analyzing the resolution and other aspects of the signal fun. The odd fields and even fields would have totally different frequency responses on the CRT device (assuming the apertures (size and shape of scanning spot) to the the same on both).

I think a projection system where all the pixels in the image could be set to their proper value all at once rather than being set in a top-to-bottom fashion, would be better than either an interlaced or even a progressive system. A memory-mapped system in which each element instantly becomes the proper color (and holds that color) based on the contents of it's mapped memory address would allow screen refreshes as fast as the memory could be refreshed by the underlying software. The frame changes could then appear like film, except without the shutter flicker and loss of light due to the shutter mechanism having to block the light while the film is traveling.

quote:

---

Evans, you seem to fixate on the beat pattern problem in scan line based systems. While I agree that this is an important aspect of resolution, it is not a be-all and end all-measure of quality.

---

Well, beat patterns and aliasing are new issues that come up when discussing digital imaging systems that you don't have to worry about when working with film with a random grain distribution. I emphasize it because 1. The topic is very interesting to me personally, and 2. I feel that it's a topic that many people overlook. I admit that I did not understand the Kell factor until today, when I actually read the only book I had that discussed it in detail, and did my own experiment. I learned something that I didn't previously know, even after years of working with digital images, and wanted to spread it to other people here.

The reason you're not bothered by a lot of beat patterns when you watch TV is that most scenes that occur do not have regular patterns with frequencies between 120 and 240 pairs of light/dark lines of resolution that occur over a significant area of the picture. Like you said, some suits and shirts have such patterns. Aperture effects in the cameras help to eliminate some of these effects since the camera's scan lines may have overlap and produce some blurring, just as aperture effects cause blurring on a monitor in some cases.

I didn't say that these issues are then be-all, end-all measures of quality, but they're issues that must be taken into account and dealt with any time that images are resampled to different resolutions during the process of doing CGI work or converting camera resolution to film recorder resoluton, converting to HDTV, or converting to NTSC.

Another interesting topic is subjective appearance. The mathematically better approach may not produce the most pleasing effect to the human eye and brain.

No need to apologize; you were trying to help after all.

If the web were the only place I had seen that explanation, I would not have given it much credibility. But I swear I've seen that same sort of thing in books and magazines for years -- and it always confused me.

quote:

---

If the web were the only place I had seen that explanation, I would not have given it much credibility. But I swear I've seen that same sort of thing in books and magazines for years -- and it always confused me.

Anyway, thanks to everyone for helping to clear this up. It's always bugged me.

---

I've seen these figures too and they'd always bugged me, and I finally wanted to do some research into the topic and understand it myself. That's why I went into such long-winded posts about it, and is why I made some test patterns.

I am employed at an academic institution on the research side, and I'm also working on a Ph.D. in Computer Science, and have a fairly strong mathematical background (B.S. degree, 1987) and feel that if I don't understand something well enough to explain it mathematically, then I don't understand it. Doing that research two days ago introduced me to something I'd never thought about before. Surprisingly, out of all the graphics and image processing books I have, only one discusses the Kell factor in enough detail to understand it. So, I played with it, and made test patterns, and was surprised by what I saw in the test patterns. I knew that beat patterns would appear if I reduced the resolution by a few pixels, but wasn't expecting the same thing when I increased it.

I totally agree, and if there is one area where the variations in resolution need to be taken into account, it is during copying and resampling. Just as the major syndicated tv shows were smart to film with 35mm and supply reduction prints to 16mm, anyone in production today is probably smart to continue using 35mm or larger random grain based source material. Jumping to digital at that end might be cheaper, but as digital resolutions change and (possibly) increase, later use of those early digital sources will become problematic.

Even though I have pretty much retired my enlarger in favor of scanned negatives and Photoshop, I still shoot 35mm color neg for the detail and (touch wood) relative permanance. Eventually, I may buy a digital camera for everyday use, but I'm not rushing out to pay a lot for something that won't give the flexibility or resolution that I want.

OTOH, what I have found useful and somewhat fascinating in concept is using a camcorder and a Snappy multiple frame grabber to generate stills. I find the idea of sampling multiple frames to get an image that is free of noise and has more resolution a fantastic idea. I am surprised that the technology hasn't been used on some of the early films where only poor prints are available.

The idea is based on the classic best two out of three or three out of five concept. In multiple frames of the same subject, any image data in a perfect sequence will be identical. If one of those images has noise, it will be quite different. By ignoring data that goes beyond a set threshold, then averaging (taking the mean of?) the remaining image information, a single image can be formed that is far superior to the original images.

In old films, that image could be used as the master image for a scene. Moving objects are a little more complicated, but do-able, to the point that some early footage could probably be brought up to close to modern day standards.

The result would be lacking in grain, so a certain amout of grain noise would have to be re-introduced to give an authentic look.

Every technology has its own weaknesses and strengths. Sorry if it sounded like I was busting your chops on resolution pairs. I just wanted to bring the issue into perspective.

quote:

---

The idea is based on the classic best two out of three or three out of five concept. In multiple frames of the same subject, any image data in a perfect sequence will be identical. If one of those images has noise, it will be quite different. By ignoring data that goes beyond a set threshold, then averaging (taking the mean of?) the remaining image information, a single image can be formed that is far superior to the original images.

---

We covered this "frame averaging" technique in the image processing class I took earlier in the year! Basically, assume that a noisy image g(x,y) is formed by adding noise n(x,y) to an original (unavailable) image f(x,y). All we have is g(x,y), and

g(x,y) = f(x,y) + n(x,y)

If multiple images are available that were all derived from f(x,y), but with different distributions of noise, then averaging the images will reduce the magnitude of the noise in the average by the square root of the number of images averaged. (Take the standard deviation!) That is, by averaging 4 images, the noise magnitude can reduced by a factor of 2.

This technique is useful if all captured images are of the same thing, and in that case, works even all the images are noisy (with differing random noise). A side effect of this technique is a slight reduction in contrast due to a slight increase in black level and reduction in white level. My professor who taught the image processing course disagreed with me when I brought this point up, but I still say that such a slight degradation in contrast will occur since if the original scene has totally black and totally white areas, and became corrupted with noise, values in the middle of the scale are equally likely to be increased or decreased due to noise ( averaging out to a net change of 0), but low values near 0 are more likely to be increased, since values below 0 are not encoded, and values near 255 are more likely to be decreased, since values above 255 will not be encoded. Therefore, black areas will be more gray in the averaged image, and white areas will be slightly darker in the averaged image, while values in-between will be closer to their true value. In real-life images where the noise is of small magnitude, it will not be a problem, but if the original images have many values near black and/or near white, and noise is great, it could be.

This method was first proposed by Kohler and Woods in 1963 (see paper reference below).

References:

Gonzales, Rafael C. and Woods, Richard E. [1992], Digital Image Processing, Addison-Wesley, pp. 187-188. -- a standard textbook used in many graduate-level image processing classes at universities.

Kohler, R. J. and Howell, H. K., [1963], "Photographic Image Enhancement by Superposition of Multiple Images", [i]Photogr. Sci. Eng., vol. 7, no. 4, pp. 241-245. -- the original paper.

What you seem to be describing is a simple averaging, which is inferior to a threshold limited averaging. Crude example:

On a scale of 0 to 256 for brightness, we measure the single corresponding pixel on 5 noisey images. They are

0
100
95
110
15

We find the "best three" numbers that are closest together -
in this case 100, 95, and 110 (This example doesn't account for surrounding pixel values)
We then find the max variation between the best three, which is 15
We then arbitrarily double that to set the threshold to 30.
Anything outside of the aprox average of the best 3 (102) by 15 points or more is discarded as being too corrupted by noise to contain useful data.
In this case, the image comes out at that 102 value.

In a straight averaging of all images, The noise would cause that value to drop well below the lowest reasonable value, 95, resulting in image degradation.