Compression Methods in Video Technology
by Luzian Linke
Data rates have to be reduced in order to transmit video data over networks. How does this work?
Enormous data streams are generated when video signals are transmitted. A CCIR-601 signal, for instance, needs 270 MBit/s purely for the video data stream this bandwidth is not even provided by high-speed networks, however the current Ethernet technology lies at 100 MBit/s. Different compression methods can be used to reduce data rates so that the available bandwidth is used optimally. A crucial factor is the method and for which application the respective compression algorithm or codec (encoding and decoding) was programmed that is used to compress or decode the uncompressed video data. Distinction is drawn here between lossless and lossy data compression. Lossless compression can be compared with an air pump: all molecules are still present during compression; they occupy merely a smaller volume. When the pressure reduces they reoccupy the original space. Many video signals contain redundant information that can be combined without any quality loss. Information is not lost in this way, however the compression factor lies only at 2:1 to 3:1.
Lossless and lossy compression
Lossy compression removes information that is unimportant for observing, and it achieves higher compression factors. Specification is made in the applicable codec of that information which is relevant and that which is not. A codec can compress the data within single frames (intraframe), between several frames (interframe) or even use both methods together. The combination and sequence of different techniques for compressing single frames decide the compression factor, compression quality and memory requirement.
Intraframe coding
(Within the single frame)
- Compression factor relatively low (15-20)
- Application: JPEG, Wavelet
Interframe coding
(Between several single frames)
- Only changes to the frame are transmitted
- Higher compression factor (70-80)
- Application: H.261, H.263, H.323, MPEG-2/4
Application purpose is crucial
Compression algorithms were designed and standardised for the different ranges of applications on the international level. Therefore they are not very suitable as alternatives to one another as they were designed for different purposes and tailored for specific fields of application. However, all compression algorithms presented here cover the streaming-applications spectrum and are established on one of the international standards. Furthermore, some of them provide enough scope to allow matching to the needs of security technology to be optimised.
ML-JPEG
The ML-JPEG compression method is an enhancement of the M-JPEG standard that was developed especially for security applications. By this method the single frames of a video sequence are also compressed using the JPEG method and frame interdependencies are not used for compression. However, new algorithms allow data quantities to be considerably reduced with ML-JPEG without any losses in quality: in the past, data quantities of 15 Kb/frame and over at half resolution were the rule in the M-JPEG method, optimisation of the compression algorithm with ML-JPEG can reduce the data quantity to 2-4 Kb/frame. 18 Kb is taken up by each frame at the top quality level i.e. at full resolution of 720 x 576 pixels.
MPEG
MPEG is the abbreviation for the compressions standard of the Moving Picture Experts Group (ISO commission) which includes simplifying or removing less-visible samples (lossy compression) and the remaining data can be recompressed by reducing identical samples (lossless compression). In addition to the JPEG compression method only the changes from the previous frame are stored (unlike M-JPEG): However, the MPEG format stores so-called intra-frames (I-frames) at regular intervals of typically twelve frames; they are single frames compressed with JPEG. The frames between these I-frames are not fully stored whenever possible. Rather, MPEG stores how moving parts from previous or subsequent frames can restore them. Anticipatory predicted frames and B-frames (bi-directional frames) are also used for this purpose. The deviation remaining per frame is still stored encoded in JPEG, though, as this method never functions perfectly.
- I-ntra frame: digitalised full frame, basis for calculating the P- and B-frames
- P-redicted frames: are predicted from previous B- and I-frames, approximately 50% of the I-frames memory requirement
- B-frames: are derived from previous B- and subsequent P-frames approximately 20% of the I-frames memory requirement (B-frames = Bi-directional predicted frames)
Their upward compatibility is an important feature of all MPEG formats. Advanced codecs can therefore use each of the older formats from the same edition:
- MPEG-1 (1992) for CD-ROM/Video-CD, MP3, 1 MBit/s 1.5 MBit/s
- MPEG-2 (1995) for transmit quality (PAL/NTSC/HDTV), 2 MBit/s - 80 MBit/s
- MPEG-3 was originally intended for HDTV and was merged into MPEG-2
- MPEG-4 (1998) for interactive video & multimedia, <64 KBit/s - 4 MBit/s