printer friendly version

Image compression processes

Katharina Geutebrück

You will remember I mentioned earlier in this series that the main differences between control centre systems relate to the picture compression process used, and the differences depend not only on what standard applies, but even how the standard was implemented. Here we look in more detail as to what these differences mean.

Specialist magazines and marketing documents from video system manufacturers have set out endless arguments for one or other compression process. Currently the widely used H.264 process is regarded as the most efficient video compression process for moving images, because, with suitable computer resources, it generates the lowest bandwidth for a given image resolution and quality.

The best compression process

However, one important point is often missing from debates about compression. The most efficient way of saving bandwidth and storage capacity, is simply not generating, transmitting or storing images or extra resolution in the first place if it is not required. So when selecting a suitable system here too we find the earlier question is relevant again: what (image) data needs to be available in what resolution and in what quality and when?

Let us imagine a control centre workstation where there are usually four images on a medium-sized monitor. Since the monitor itself only has Full-HD resolution, each of these images is displayed with a resolution comparable to D1. Yet these pictures are sent from the camera through the network in full-HD resolution, even though they cannot be displayed in this resolution.

However, with the right forward planning and more intelligent processing, the situation could be very different.

Imagine the same workstation using a different implementation of H.264, this time each of the four cameras is sending D1 resolution pictures. Something happens in one of the camera scenes which the observer wants to see more closely. He selects full screen display, and the monitor shows this camera image now in full resolution (full HD). In other words, as soon as the event ‘operator action: switch this camera to full picture mode’ is generated, this camera immediately switches to full-HD streaming and delivers the appropriate high resolution images. At the same instant, a signal is sent to the other three cameras so that they cease transmitting images over the network. (Note that if the camera does not switch immediately but continues sending D1 resolution pictures then the image would look pixelated so the camera reaction time is important.)

With intelligent systems like this, the only bandwidth required is for one full-HD (or 4 D) stream. Of course, the camera or the central system also has to have the ability to provide different, independent streams in different qualities to allow the quality and picture rate for recording to be decoupled from that required for viewing. Otherwise with this set-up the quality of the recorded pictures would vary and may be rendered unsuitable for later viewing purposes.

Peculiarities of H.264

Having originated in the world of multimedia technology, the H.264 standard in its standard form brings with it some properties which are disadvantageous for security applications. In order to achieve the highest possible compression efficiency, whole picture sequences are analysed, compressed and consolidated as so-called groups of pictures (GoPs). Included in the data stream for nearly every picture are references to the previous picture and also to the following one – thus effectively chaining the pictures together. Except for the so-called I-Frame, the first picture in the sequence which is compressed independently of the rest of the pictures and can therefore be displayed independently, all the other data from the previous and following pictures in the GoP is needed by the display computer in order to display any of the other pictures.

What does this mean for the user?

* Image artefacts from data loss: In a network it is not unusual for individual data packages to go missing. In the ‘chained’ process described above, any such loss means the remainder of the GoP cannot be displayed correctly.

* High computation loads during decompression: The need to decompress the data from many images, before even one can be displayed means that the standard display computer has to work flat out to display four camera videos with full picture rate (25 fps). If a further camera is switched then all the streams hang up and fluent replay is no longer possible.

* Jerky display in fast forward or backward replay: When stored images are replayed to analyse details of the recorded event, operators need to use fast forward and backward functions as well as frame by frame forwards and backwards.

* Inaccurate picture search: When responding to searches for images relating to a specified event or to a particular date and time, many systems only provide the nearest I-frame, as these are the only ones which are saved along with metadata. Their manufacturers regard these as representative of the complete picture sequence, but depending on the length of the sequence, the relevant image may be several seconds away.

The solution to these problems lies in specially developed variants of the H.264 standard which avoid the chain problem. These still conform to the standard, offer a slightly lower compression efficiency, but provide much more convenience and other significant advantages. They are known by their proprietary names of ‘P-chain-free’ and H264CCTV.

Share this article:

Further reading: