A method and apparatus for achieving a high compression of a video signal. The PEL-by-PEL difference between an input signal consisting of digital PEL values of a scanned video signal and a motion compensated interframe prediction signal is decomposed into several narrow bands using separable two-dimensional quadrature mirror filtering (105). Each sub-band is quantized by a symmetric uniform quantizer (106-1-106-4) with a center dead zone. Entropy coders (124-1-124-4) code the quantized values by variable word-length coding the nonzero quantized values and transmitting that information with the corresponding run-length coded positional information. The outputs of the coders are combined by a multiplexer (108), the output of which is input to a buffer (123) which produces a constant rate bit stream. When the occupancy level of the buffer approaches maximum capacity, the dead zones and step sizes of the quantizers are adjusted to force more zero value quantized levels thereby reducing the amount of data being input to the buffer.