Spatio-temporal subsampling and transform coding of HDTV signals

Abstract

A two-stage coding system for reducing the usually high data rate of an HDTV signal to less than 140 Mbit/s is herein described. Three-dimensional subsampling is used to reduce the number of samples of the digitized interlaced source signal. A motion-adaptive filter structure adjusts the three-dimensional spectrum of the television signal to some reduced region which is supported by the quincunx sampling pattern. The visibility of the transition between different spatial resolutions is decreased by 3D-filtering of slowly moving areas. As a result of subsampling, the sampling frequency is halved. Transform coding of the remaining samples is then performed. The quincunx sampling structure is rotated within blocks of 8 by 8 to result in a rectangular block structure. Different possibilities for transforming the quincunx sampled field have been investigated and compared in terms of energy concentration, entropy calculation and coding efficiency. A data reduction of the transform coefficients is sought in the range of four and five. A modified threshold coding algorithm is used to code the coefficients. Sampling, normalization and quantization are adapted and controlled by the buffer status. The buffer equalizes the variable bit-rate at the output of the variable length coder. All steps of signal processing have to be adapted to preserve the high quality of the original signal.