amirm

Well, I typed real slow. It is the darn forum software which outputs it so fast . Thanks .

Ah, I see what the issue is. You are actually trying to understand this stuff . So let me expand a bit. There are actually two definitions here, both with the same name but each mean a different thing! The first should more correctly be called, "TV Lines per picture height" but unfortunatley, people omit everything after "TV Lines." This is the one I described above in my post. The motivation for this definition was to make sure that if a circle is drawn at 1 inch radius, that it has the same number of pixels in horizontal and vertical dimensions. Since NTSC pixels are rectangular by the 4:3 ratio, we use that math to arrive at the normalized horizontal resolution to go with our vertical resolution (which never changes in NTSC). In this kind of conversation, no one really cares about how many lines this actually captures from real life. We arrive at a number and use that to compare equipment. What you seek is the true definition of "TV line." That, the exact resolution one can capture out of the system. Alas, this is a tricky problem to solve because it is a subjective discussion, not objective. Now if you read my long post, I talked about something called a Kell factor. That is an attempt to answer your question. That is, if I have 10 pixels of resolution, how many pixels of real life can I capture? The answer may seem simple but it is not. Imagine you have a series of alternating white and block dots you are trying to capture. I will represent White with X and Black with O. So assume the following simple example of the source: X O X O X Now let's assume we have a camera with 3 pixels worth of resolution. For the sake of simplicity, let's position these pixels precisely where the white dots are (each camera sensor pixel is shown as "C"): X O X O X <--- image you are trying to capture C C C <--- camera pixels In this scenario, the camera sees only the Xs and outputs all white: X X X <---- output It is obvious that we lost the original image and that we need more than 3 pixels to detect three white dots seperated with blacks. Now, if we had an extra sensor pixel in between, that is, 6 pixels total, we could capture the true transitions from black to white. This leads us to think that the true resolution of such a system is half as much. But this would be too pessimistic view of the world. Why? Because the pixels don't have to match the way I showed. And that if the subject moves, and/or pixels don't align, then we are likely to pick up the black and white pixels. An RCA engineer in 1930s by the name of Kell studied this effect and came up with percentage to use here that is more optimistic than the 50% shown in my example. Based on his work, the number 0.7 is used as the effective resolution for CRT systems. That is what I used in my math in the long thread to arrive at the true vertical resolution. That is, you multiple your pixesl by the Kell factor to arrive at the subjective resolution of the system. Of course, there is no telling what this number should be since subjects vary as do sensor sizes. Kell himself went from 0.64 I think to 0.85 and others have come up with different numbers. Net, net, if you are shopping for stuff and want to equalize specs then use the first definition because that is what the industry uses. The number won't translate into real life resolution but maybe we don't care to actually measure the resolution .

Simple: there are two different measures here: one is the number of pixels and the other is "TV lines." The two are related but through a complex chain which I described in the lengthy post. Quick summary is this: To get horizontal TV lines from X number of pixels, first you multiple X by 3 and then divide by 4 to compensate for the fact that NTSC uses rectangular pixels which have a 4:3 aspect ratio. So your CIF horizontal pixel resolution of 352 just shrank to 264 TV Lines. This is lower than 330 TV lines which NTSC can carry at the limit. In other words, the TV line metric derates the horizontal resolution as to normalize it to the vertical resolution. Hope this makes it more clear.

Appreciate the kind words . I know very little about video used in CCTV applications so am learning a lot here. And thought I pass on the bits I know. Actually, that assumption is not correct. NTSC does exceed the resolution of CIF. As I said: "And in the case of a DVR doing a poor job of compressing the content, then the extra resolution NTSC has over CIF is probably lost," it is the combination of poor video compression and NTSC that might eliminate the extra resolution benefits of 4CIF. Put another way, if you have a static shot with a lot of bandwidth, then 4CIF will look better. But once the image moves, and you introduce compression artifacts, then much of that resolution might be lost. As a way of example, I was at some IP camera company web site showing how they could easily capture license plate of a car and showed the difference between their megapixel solution and that of analog camera. Then I go look at their live feed of the parking lot, and the license plate is a complete blur with blocking artifacts of MPEG-4! So one needs to look at these platforms as complete systems end to end and evaluate that. Unfortunately, despite research for a while I have not found anyone benchmarking these cameras side by side with all of these considerations thrown in. It is not an easy test but a necessary one to gather good data. The camera portion certainly goes beyond what NTSC can handle. Per above though, that is not the real explanation. However, there is one topic I did not cover there that I will here. Namely, if you take a higher resolution sensor, and then filter it down to what NTSC can handle, you get some natural noise reduction out of it. In layman terms, assume that we average pixels to arrive at fewer ones to transmit. If average a noise pixel with a black one, then the noise level goes down by a factor of two. So in some sense, an ultra high resolution sensor helps produce a cleaner picture.

Adding on to my previous post, what of the things H.264 does well over MPEG-4 (part 2), is that it degrades much more gracefully when pushed. A lot of detail gets lost in a sea of compression artifacts when things move, or there is picture noise. H.264 deals with this much more effectively (assuming all the features are implemented). So in some sense, the dynamic resolution of H.264 is higher than that of MPEG-4.

I can certainly buy that to some extent. But that is a different argument than it being "near DVD." Composite video has nothing close to DVD resolution given the limitations of NTSC modulation. So it is not surprising that a CIF encoding doesn’t lose much in resolution in that scenario. And in the case of a DVR doing a poor job of compressing the content, then the extra resolution NTSC has over CIF is probably lost. Per my post here: http://www.cctvforum.com/viewtopic.php?t=10415&postdays=0&postorder=asc&start=45, (see middle of the page), 540 TVL refers to sensor resolution, not what comes out of the camera. The output of the camera is limited to NTSC spec which is 330 lines. So good to hear that your experience matches theory . That is a correct description of the end results. Technical explanation is that the DVR NTSC decoder, demodulates the source and separates the signal into YUV components (i.e. black and white and color). It then digitizes these samples at D1 resolution if told to, but that source being NTSC, was filtered at the camera end to 330 lines. So even though the DVR input logic attempts to sample the source at higher resolution (fill the buffer as you state), the samples do not change at higher frequency than 330 lines. Well said.

CIF images are going to be soft no matter how little compression artifacts there are. Sure, if you are bandwidth starved, then an H.264 at lower resolution may be superior but if motion is small, even that won't be the case. How do you get near DVD when you have one quarter the resolution? I am not sure that says anything . Apple's profile of H.264 is the lower complexity one anyway. The highest fidelity is HP which is used in production of Blu-ray. The issue here is that standards do not specify encoder performance. Two H.264 could be wildly different from each other. With H.264 specially, there are a ton of shortcuts that can be made to achieve high frame rate. Software solutions running in DSP/PCs are liable to take the most shortcuts. Best solutions are used in broadcast products and even there, they take shortcuts relative to PC software professional encoders. Broadcom has nice H.264 encoder silicon but I am not sure any DVR vendor uses it yet.

That is a good point in that PC based recoding goes up to broadcast standard. Which is quite nice. Sadly, the signal arriving at the card from the camera, doesn't approach those numbers unless you use the S-video (Y/C) cable. You lost me there . As long as you are using a single coax to connect to PC DVR, you can't get those numbers (and at any rate, TV Line is not the same as the resolution numbers used above for DVR). The signal is filtered (lowered in resolution) at the camera end before being put on that cable due to limitation of NTSC broadcast standard we live in. The numbers you mention per my post, are the resolution of the front-end/sensor of the camera, not what you get at your DVR -- PC or otherwise.

No question Rory will make this conversation interesting. Now, who is Rory? New kid on the block.... Amir

The CCD sensor can have any resolution. You can even have 1000 lines of resolution and subsample down. But per my above note, your limit in non-IP cameras, is that of NTSC modulation. That is the weakest link. And that is well below 420 lines. An IP camera on the other hand, is able to transmit resolution well above that since it is not bound by NTSC standard. However, in the process of compression, it adds non-linear distortion which can bring its effective resolution well below the sensor.

While your statement is generally correct, in case of a DVR, it also degrades the image in the process of compressing the image. And there, we do have an impact due to higher level of grain/noise, which reduces efficiency of compression, causing more artifacts. The level of redundancy in the image goes down as you replace real life images with a lot of noise. But yes, the source of the problem is the camera/lighting....

Ah, finally a subject I can contribute to! Have been reading this forum and learning from it for a while. Thought it made sense to register now and provide a mini-tutorial on the subject of resolution. By way of background, I an engineer with two decades of experience in analog/digital video (plus bunch more in other areas). So if some of this doesn't make sense, feel free to ask questions. At high level, one can make any of the claims people make and be both right and wrong! The key thing to understand is that this is a rather complex situation. We have a digital imaging sensor paired up with analog transmission over a coax cable. One has to understand both components to know the full resolution of the system. So let’s start with the easier, digital part first. The sensor today is a digital device meaning it has a specific resolution. The highest fidelity sensor for standard definition (SD) signal meant for NTSC transmission in US would have a resolution of 720x486 (some use 720x485 – don’t worry about the difference). This means that the sensor resolves 720 pixels horizontally and 486 vertically. This by the way, is the resolution of DVD if you watch it over a digital interface (e.g. HDMI/DVI) connected to a digital TV (LCD, etc.). SD television is a “4:3â€