MPEG VIDEO OVERVIEW

GenXTechno Tags: mpeg, motion, picture, expert, group, video, compression

Although the MPEG-1 standard is quite flexible, the basic algorithms have been tuned to work well at data rates from 1 to 1.5 Mbps, at resolutions of about 350 by 250 Pixels at picture rates of up to 25 or 30 pictures per second. MPEG-1 codes progressively-scanned images and does not recognized the concept of interlace, interlaced source video must be converted to a non-interlace format prior to encoding. The format of the coded video allows forward play and pause, typical coding and decoding methods allow random access, fast forward and reverse play also, the requirements for these functions are very much application dependent and different encoding techniques will include varying levels of flexibility to account for these functions. Compression of the digitized video comes from the use of several techniques: Sub sampling of the chroma information to match the human visual system, differential coding to exploit spatial redundancy, motion compensation to exploit temporal redundancy, Discrete Cosine Transform (DCT) to match typical image statistics, quantization, variable length coding, entropy coding and use of interpolated pictures.
Algorithm Structure and Terminology
The MPEG hierarchy is arranged into layers (Figure 1).

This layered structure is designed for flexibility and management efficiency, each layer is intended to support a specific function i.e. the sequence layer specifies sequence parameters such as picture size, aspect ratio, picture rate, bit rate etcetera , whereas the picture layer defines parameters such as the temporal reference and picture type. This layered structure improves ro-bustness and reduces susceptibility to data corruption.
For convenience of coding, macroblocks are divided into six blocks of component Pixels four luma and two chroma (Cr and Cb) (Figure 2).

Blocks are the basic coding unit and the DCT is applied at this block level. Each block contains 64 component Pixels arranged in an 8x8 array (Figure 3).

There are four picture types : I pictures or INTRA pictures, which are coded without reference to any other pictures; P pictures or PREDICTED pictures which are coded using motion compensation from previous picture; B pictures or BIDIRECTIONALLY predicted pictures which are coded using interpolation from a previous and a future picture and D pictures or DC pictures in which only the low frequency component is coded and which are only intended for fast forward search mode. B and P pictures are often called Inter pictures. Some other terminology that is often used are the terms M and N, M+1 represents the number of frames between successive I and P pictures whereas N+1 represents the number of frames between successive I pictures. M and N can be varied according to different applications and requirements such as fast random access.
A typical coding scheme will contain a mix of I,P and B pictures. A typical scheme will have an I picture every 10 to 15 pictures and two B pictures between succesive I and P pictures (Figure 4).

Prediction (P Frame)
The predicted picture is the previous picture modified by motion compensation. Motion vectors are calculated for each macroblock. The motion vector is applied to all four luminance blocks in the macro block. The motion vector for both chrominance blocks is calculated from the luma vector. This technique relies upon the assumption that within a macroblock the difference between successive pictures can be represented simply as a vector transform (i.e. there is very little difference between successive pictures, the key difference being in position of the Pixels) (Figure 5).

Interpolation ( I Frame)
Interpolation (or bidirectional prediction) generates high compression in that the picture is represented simply as an interpolation between the past and future I or P pictures (again this is performed on a Pictures are not transmitted in display order but in the order in which the decoder requires them to decode the bitstream (the decoder must of course have the reference picture(s) before any interpolated or predicted pictures can be decoded).