Our Man in the Field: The Process of the IP Solution, Part X

Charlie takes an in-depth look at the common IP-video compression schemes

MPEG works on the principle of using three basic structure pieces. We have the "Intra Frames" which are referred to as the “I” frames. These are whole, uncompressed images of the whole scene. Next up are the "Bi-Linear" or "B" frames. These are recorded images of those objects in the scene that change ... that is to say, what moves. Lastly, we have the "P" frames. These frames are used to do comparisons between the preceding I-frame and subsequent B-frames.

OK, so let's try and put this complicated little puppy into action. The amount of compression will be determined by the number of B-frames that are interjected between each I-frame. The P-frame is injected at the end of the B-frame string and looks back to the previous I-frame and jots or pushes the objects of the recorded B-frames back into position as based upon their actual location in the overall scene. Despite the complicated sound of the process, MPEG can be an extremely efficient form of compression. File sizes will range from one point four (1.4) to twelve (12) kb.

However, here are the potential problems with MPEG video. First, there is no time, order or date stamp applied to the B-frames. Therefore, it is conceivable that I could take an image of you walking along the sidewalk and plant you into a video stream. No one would be able to prove it one way or the other. You would not appear in the I-frames, but what the heck. The bottom line is that in some court cases, only the I-frames would be allowed for viewing by a jury, as they are the only pieces of information that can be proved as whole or un-tampered. The second problem is that if a B-frame becomes corrupt and you are working with heavy compression (more B-frames), the corruption will remain and worsen until the next I-frame erases the corruption. So we end up with the famous car rolling down the street when it breaks in half and the rear portion of the car remains on the screen with other cars driving through it. This is caused by image memory, also known as bleed over. So, the more compression, the more room for corruption ... it sounds like city politics, eh?

H.263, H.263+, H.263++ are all variations of the original standard for video telephony or video conferencing via the internet. It starts by processing images in much the same way as JPEG, but then it adds the twist of "Motion Compensation Prediction". This means that is sees an object moving and literally maps its position into future frames in "anticipation" of the vehicle reaching the next point. This makes H2.63 very fast. It also makes it very susceptible to problems with sudden movements. Granted a car is fairly easy to predict in most cases. However a human waving her arms around in a frantic flurry may cause the engine to fail in its ability to predetermine the position of the hands and such, and so drop them from the recorded, compressed image completely. In actuality, the background shows through the fluttering hands so they appear to disappear. File sizes range between point five four (.54) to eight point seventy seven (8.77) kb.

The latest and greatest compression schemes are of a truly digital format. The designers are literally converting the visual information into a form of text ... pixel by pixel. I know this seems odd, but think about it. I literally write a picture. Pixel one is red, pixel two is green, pixel three is blue, etcetera. I can store hundreds or thousands of pages of text into the same space as it takes me to store fifty images. The net result is that the creators of these compression schemes are claiming 2,400:1 compression factors with up to two (2) CIF playback. I realize that I have over simplified the process, but not by much. I have basically told you that you can cut your two or three terabytes storage requirement to one hundred gigs or less. The problem is that these systems are extremely proprietary, so you may have to supply some translation when your surveillance video appears in a court room.

OK, I think that I have proven my case that all compression schemes are crap (as was my tenet in the last column). Anytime you take something that offers the requirements of detail and you compress it, you must give up something. So what do you do? You can't afford to work without some format of compression. You work with the compression scheme that best suits your application. You determine which one of the compression schemes is best for you through, on-site demonstration and testing. You cannot go to a show and pick the one that is best for you unless you have a fairly standard application with few or no surprises. Buyer beware.