This report is also available as an Acrobat file.
4. COMPRESSION STANDARDS
Uncompressed PAL video as a digital signal needs 140 to 270 Mbps, while uncompressed digital HDTV needs 1.2 Gbps. A digitised colour picture at 35 mm film resolution needs about 80 Mbyte. One minute of 8 bit sound sampled at 22 kHz needs 1.3 Mbyte. These requirements are large for storage and transmission in computer networks. Fortunately many applications can live with video or sound that is not perfect. It is therefore possible to produce coding schemes that will drop certain information, or make a guess at likely values. Such coding can result in
compression ratios of up to 200 times, so PAL video can be compressed to 1.5 Mbps. Compression must be done to a standard to enable decoding. However their are many compression methods. Microsoft Video for Windows, Quicktime are proprietary examples. A standard called JPEG is used for still colour images. For video conferencing a standard called H.261 which is part of the H.320 standards family is used. Further video standards are MPEG-1 and MPEG-2 which gives a higher quality picture to MPEG-1, but at a higher data rate. MPEG is likely to be important for video on demand. An MPEG decoder card is available for about £300. Compression of video to MPEG can now be done in real time, but the hardware to support this is expensive.
4.1. JPEG Compression
The JPEG (Joint Picture Experts Group) [ISO10] is the first international digital compression standard for multi-level continuous tone still, black and white or colour images. It typically compresses images to 1/10 or 1/50 of their original size. It is based on use of a discrete cosine transform and requires same level of processing to compress and decompress an image. JPEG aims are :-
Most implementations have been only of sequential encoding. A 10 MHz JPEG chip can typically compress a full page 24 bit colour 300 dpi image from 25 Mbyte to 1 Mbyte in about one second. JPEG takes each lock of 8 x 8 source image samples and codes them into coefficients. The most important coefficients are then preserved. JPEG compression and decompression by different systems are not guaranteed to be the same, but a an accuracy test is available.
- To be applicable to any kind of continuous tone digital source image.
- To be able to be implemented in hardware or software at reasonable cost.
- To support the following modes of operation :
- Sequential encoding, i.e. left to right and top to bottom scanning.
- Progressive encoding, using multiple scans so that the image builds up gradually.
- Lossless encoding, in which the compresses image can be decompressed to be identical to the original.
- Hierarchical encoding in which the image is encoded at multiple resolutions, so that lower resolution displays can be accessed without having to decompress the full resolution image.
JPEG has been used for full motion video by compressing each frame of the video. For a 640 x 480 pixel 24 bit colour JPEG compresses each frame of about 1 Mbyte by about a factor of 50, which results in a data rate of about 5 Mbits/sec for 30 frames/sec. For this high data rate most
implementations use a small window in the PC of 256 x 240 pixels which reduces the data rate by a factor of five. Intel DVI compression does the same. Motion JPEG does not support audio compression which must be done separately.
4.2. MPEG Compression
The MPEG (Motion Picture Experts Group) [ISO11] has so far defined two compression algorithms.: MPEG-1 and MPEG-2. A common misconception is that MPEG-2 is a replacement for MPEG-1. Each algorithms has been specifically targeted at different bit rates. MPEG-2 runs at higher bit rates than MPEG-1. There are no firm constraints in either algorithm and it is possible to run MPEG-1 video at very high rates. MPEG requires more
processing power to compress video than decompress, so it is ideal for video film distribution. MPEG-1 chips on the market provide about a 200:1 compression to yield VHS quality video at 1.2 to 1.5 Mbps
The MPEG specifications allow manufacturers to implement different proprietary, but MPEG compliant algorithms. There is therefore no guarantee the the output quality of MPEG encoders will be the same. Basically the user pays for what they see. MPEG takes advantage of temporal redundancy in video pictures by specify three type of pictures:-
Pictures may not be sent in the order in which they are displayed, if reference pictures are needed for reconstruction. MPEG also provides for synchronisation of audio and video streams.
- Intra Pictures or I-pictures are coded using only the information present in the picture itself using cosine transforms. I-pictures use about two bits per coded pixel and are used about every two seconds.
- Predicted Pictures or P-pictures are coded with respect to the nearest previous P or I-picture and use forward prediction of the video picture content.
- Bi-directional or B-pictures that use a past and future picture as a reference. B-pictures provide the most compression and average out noise. Typically two B-pictures will separate a P-picture.
Motion compensation is a technique used to enhance the compression of P and B-pictures by examining the spatial difference between pixel blocks within the picture.
I B B P B B P
Video Stream Order
I P B B P B B
Stream Versus Display Ordering
MPEG-2 uses many similar techniques, and allows for many different resolutions and frame rates. It also takes advantage of motion prediction between video fields, enabling higher compression ratios. Both MPEG-1 and MPEG-2 can run at reduced resolutions by reducing the number of pixels to be encoded by a factor of two before compression.
There are optimal choices for the use of MPEG-1 and MPEG-2. Below link rates of 3.5 Mbps MPEG-1 provides better video quality (VHS standard) if the pixel input is reduced as described above. However for data rates above 5.0 Mbps MPEG-2 provides better quality, Super VHS or above video.
Virtual Environments Visualisation