MPEG background

Technical info about MPEG and MPEG files.

The name MPEG does not stand for a compression format but stands for Moving Picture Experts Group. It is a working group of the international standards organizations ISO/IEC responsible for the development of standards for digital audio and video encoding. Their home page can be found at http://www.chiariglione.org/mpeg/. The group was established in 1988 and has generated several standards for digital video compression which simply carry the name of the group. Accepted standards developed by the group include MPEG-1, MPEG-2 and MPEG-4. Currently, the group is working on a new standard, MPEG-7.

History of MPEG standards

MPEG-1

This format was accepted as international standard in 1992. It was developed to be able to store video and audio on digital media. One of the important aspects was that, for uninterrupted play back, the streams could be decoded real time by the processors that were available at that time. Although the standard itself allows use of any bit rate, it was developed for transfer speeds of around 1500 kb/s for direct playback from CD. The standard was developed to store video with either 625 or 525 lines. Only progressive (no interlaced) video can be stored in MPEG-1. The standard was very popular in gaming applications and was also used in VideoCD.

MPEG-2

As stated above MPEG-1 was developed to be handled by the processors that were available in that time. However, development of faster, more powerful processors, opened the door to higher levels of quality than the MPEG-1 standard allowed. Furthermore, there was also interest by the broadcasting industry to go digital. Analog material for play back on TV has a higher number of pixels than MPEG-1 allows and it is also interlaced. Therefore, a new standard was required. For this purpose MPEG-2 was developed. Besides the ability to handle larger frame sizes and interlaced video it also allows features like multiple audio channels, like 5.1, advanced audio techniques, like Dolby AC-3, and copy protection capabilities, like CSS.

MPEG-4

MPEG-4 is the latest accepted standard. The basic goal was not to establish a better quality compression but to add other features like interactivity. It is used for digital television and games where interactivity in the video is required. The use of the compression is subject to a paid licensing scheme, but the algorithm was hacked pretty soon after its released by others in order to be able to use it license free. The most well known hacked version of MPEG-4 is DivX.

MPEG Encoding techniques

There are basically two types of compressions used in MPEG. One is spatial compression, the other one is temporal compression.

Spatial compression

Video consists of a sequence of frames. Each frame consists of a number of pixels. Compressing an individual frame in a way that it takes less data to describe it is done with spatial compression. The mechanism is similar to the technique used to compress digital stills in JPEG compression. The compression algorithm analyzes the frame and determines what data can best be left out with as little loss of quality as possible. Subtle color differences are left out and sharp edges become blurred a little.

The compression is carried out on blocks of 8 x 8 pixels. This is the cause that heavy MPEG--compression can result in blockiness, a well known phenomenon in low quality VideoCD's.

Temporal compression

Temporal compression works by compressing sequences of frames. It is based on the fact that, of course, there is a lot of similarity between subsequent frames. Only if a new scene is started with a hard cut there can be zero similarity between two consecutive frames. The basic trick is that some of the frames are generated by interpolation of other frames. The compression algorithm analyzes a sequence of frames, determines the resemblance and stores certain frames just by storing "differences" from a nearby frame. Just as for spatial compression, temporal compression is on the level of the blocks defined by 8 x 8 pixels. The temporal compression mechanism will create three types of frames:

• Intra frames.
  I-frames are independent from any surrounding frames.
• Predicted frames
  P-frames are stored as a "delta" to a preceding I-frame.
• Bi-directional frames
  These are frames that contains "delta" information from a preceding I-frame and a succeeding P-frame.

They I-frames only contain spatial compression and therefore take a lot of space. The other types of frames only contain limited amount of information and therefore require a significant smaller amount of storage. This is the real big gain in MPEG compression.

A sequence of frames starting with an I-frame and ending with a P frame is called a Group of Pictures (or GOP). Simple encoding algorithms will create the same frame sequence throughout the whole movie. Multi pass encoding will create different GOP sequences throughout the movie: more B-frames in static scenes with a lot of similarity between successive scenes and lesser B-frames in dynamic scenes.

The interdependence of frames in an MPEG-compression also explains why MPEG-files are less suitable for digital video editing. When you want to cut the movie at a position of a P- or B-frame you will break a GOP and a lot of frames around the edited frame need to be recalculated in order to regenerate one or more GOP's which will take a lot of calculations.