Video Compression: A Second Approach

If digital video signals could be processed in such a way that they were recorded on computer hard drives, economically, without any apparent loss of quality, then the possibilities in editing, graphics and animation would be endless. In addition, if digital video could be fed into the same bandwidth as conventional analog signals, viewers could receive studio-quality images in their homes.

The techniques of Motion JPEG (Joint Photographics Experts Group) and MPEG (Moving Pictures Experts Groups) are now widely used for the creation of computer images that are recorded on discs or CD-Roms, although none offer optimal results for transmission. However, MPEG-2, a ratified ISO/IEC standard, which the industry adopted at an incredible speed, motivated almost entirely by the strong desire to provide viewers with an immense variety of direct-to-home (DTH) programs via satellite or cable, using set-top boxes, did yield very positive results.

The sole purpose of MPEG-2 is to convert the transmitted bit ranges into something more manageable. Its success is based on the compression of primary information in two areas of motion picture. The first is the information contained in each box, called spatial (it relates to space such as blue and sky); the second is detail or temporal (time-related), which does not change from frame to frame.

MPEG-2 is indeed a world standard. The system accommodates everything from compressed computer information ranges of less than 4 Mbits/sec, to conventional television of 10 15 Mbits/sec, and high-definition television operating up to 8 Mbits/sec (known as levels). MPEG-2 provides flexibility in the type of compression used for each level. These types are known as profiles and vary depending on usage, from the full 4:2:2 signal, to the removal of entire frames.

Encoders can change considerably depending on the application, so the details of the encoding scheme must be transmitted at the same time as the information, to enable the decoder to reconstruct the signal. In this way, coders can be designed to handle several levels using different profiles at the same time. Many of the 525- and 625-line streams use a main profile at a main level (MP@ML).

Temporal compression is designed to minimize duplication of information contained in successive images. This is achieved by transmitting moving vector data and other different information, rather than repeating the entire image again. To facilitate motion prediction, MPEG-2 separates the video into three types of images:

- Images I (Intracoded)

- P Images (Predictive Coded)

- Images B (Bidirectional Interpolated)

Images I are the key reference for the other two types of images. They are derived by selecting the information from a single chosen box or field (spatial compression). Still images are best preserved using full frames, but since the field range is twice the frame range, movement will be better off using field-based images. Some of the MPEG-2 encoders are able to analyze the video signal that enters them to determine the changes between successive fields. If there are no changes between the even and odd fields, the encoder presumes that the two are part of the same box and encodes them as such.

Changes between fields are noticed and converted into motion vectors that are encoded into information and then interpreted by the decoder. In this way a substantial reduction in bit ranges is achieved. The changes are transmitted as P images and B images. The P images are predicted directly from the previous I images. B images are derived using information from I images or P images and these reference sources can be in front of or behind the B images being created. Hence the term bidirectional interpolation. Both P and B type images are also spatially compressed before being transmitted. The technique of motion compensation using the above method is known as temporal compression.

All three types of images are transmitted sequentially in a group (GOP), where the first image is always type I. There are usually 12 images in a GOP, but some decoders can detect changes between successive fields, and if the change is substantial, the encoder assumes that there has been a scene modification, so it will force a new image I. This causes the sequence to start over. GOPs are sent in a video sequence with information that defines the size of the image, its ranges and its nuances of quantization. The video sequence and all its small elements provide unique start codes that make it easy to detect the set-top box.

The only flaw in generating these virtual images is that engineers have yet to find an easy way to edit B and P images. Compression ranges in the order of 25:1 are achieved by MPEG-2 and are considered as a compression format by distribution.

The goal of spatial compression is to minimize duplication of information in each image. Bit reduction is achieved first by modifying the video information of space and time at the main frequency using the discrete cosine transformation (DCT) method and then by applying quantization and variable-length encoding techniques to reduce the bit range.

The DCT (which uses a trigonometric formula derived from Fourier's theoretical analysis) is used to transform the information in each block of 8 x 8 pixels to blocks of frequency coefficients of 8 x 8. In the frequency range, the high energy (and most noticeable) of the image elements is represented by low frequencies in the upper left corner of the block, and minor details are revealed as high frequencies towards the lower right.

After encoding the DCT, the information is subject to a quantification process, to reduce information in the area of high frequencies, where the human eye is less sensitive. DC components are usually quantized at 10 bits, because if more accurate quantization is employed at low frequencies, the same blocks can begin to become visible in the images.

To create a stream of video compression bits, the frequency coefficients of 8 x 8 are zigzag scanned from the top left to the bottom right and the high-frequency areas are represented by zero currents. Information reduction can be achieved by transmitting the number of zeros instead of the usual coefficient values. The last stage in the spatial compression process employs variable length coding (VLC). The VLC assigns shorter coded words for frequently occurring events and longer coded words for less frequent events. MPEG systems use these spatial compression methods to reduce bits.

Now what? Before we can store or transmit this information, we have to mix audio, video, and information systems together. There are usually two audio/video multiplexers . One takes the elementary audio and video currents and produces a program current, and the other uses the same information to generate the current transport. Program streams are typically reserved for robust transmission paths where errors are unlikely to occur. Program stream information packets can be of different lengths and contain a relatively large number of bits. A transport current is always 188 bits in length, and is designed to be used in environments where errors are a probability.

MPEG-2 has become the international standard for video compression applied to any signal that is to be simply stored, distributed and displayed. CD-Roms, for example, are being developed with MPEG-2 compression methods. Best of all, MPEG-2 has been adopted globally as the standard compression for DTH satellite television and in the future, for cable and digital terrestrial television (DTTV), including high-definition television (HDTV).