The optimal use of I, P and B frames in MPEG-2 coding

The paper describes a study undertaken to assess the data requirements of the three different frame types of MPEG-2; `I' frames, `P' frames and `B' frames. There are many ways that a combination of these three frame types can be configured to make up a group-of-pictures (GOP) and the intention of this paper is to provide a basis upon which different GOP sequences can be compared. For example: in a long GOP typical of that used for transmission, there may be the need to form a `closed' GOP to allow clip editing. Such a closed GOP must terminate in a `P' frame rather than a `B' frame to avoid association with the `I' frame of the next GOP sequence. What is the bit-rate increase associated with this compromise? This paper provides data to allow calculations of this type to be made. The paper assumes a working knowledge of MPEG-2 coding, the standard of which is now readily available. The results used in this paper were obtained using a straightforward set of software based on two programs: a motion vector estimator generating a file of best motion vectors for each macroblock and, an encoder program performing motion vector compensation, DCT coding, quantisation, runlength coding and entropy calculation.