Bit Rate Control in Real Time Video Transmission over Atm

As we have discussed in Chapter 2, the current international standards for video coding (Motion JPEG, H.261, H.263, MPEG-1, MPEG-2 and MPEG-4) all make use of DCT-based encoding to remove spatial redundancy. For motion JPEG, each frame is partitioned into blocks of fixed size, and each block of pixels is transformed using DCT and then quantized. A block that contains little or no detail will result in a very small number of nonzero coefficients, while a block containing high detail or high spatial frequency content will produce a larger number of nonzero coefficients. Therefore, the number of bits required to code each block will vary within a frame. This bit rate variation within a frame also holds true for intra-coded framed in H.261, H.263 and MPEG standards. If we consider a motion JPEG type of encoder that uses only intraframe encoding, then the coded bit rate will vary from frame to frame depending on the complexity of each frame. A complex frame with high detail generates more bits, while a smooth frame will generate fewer bits. The variation in bit rate is usually gradual within a single "scene." In Chapter 2 we discussed video encoding standards such as H.261, H.263, and different versions of the MPEG standard. These standards all use motion prediction to remove the temporal redundancy across frames and to achieve higher compression efficiency compared to motion JPEG [74]. However, motion prediction introduces further variations in bit rate. Within a motion-predicted frame, the transmitted data consists of coded motion vectors and difference block coefficients. In general, a scene containing little or simple linear motion (e.g., a camera pan) can be predicted accurately, meaning the difference coefficients as well as the motion vectors can be encoded with a relatively small number of bits. On the other hand, a scene containing a complex motion such as a rotational motion is not easy to predict. Therefore a large number of bits