Adaptive transform coding of viewphone signals

The initial phase of the research programme involves the design, development and construction of a reliable and versatile Video Acquisition and Display System (VADS) to provide a means of subjective evaluation of the results of computer simulations. An orthogonal transformation generates a set of uncorrelated coefficients and has, in addition, an energy compaction property which leads to more efficient coding of the source data. As an introduction to its application, its use in the detection and correction of transmission errors is considered. Simulations are performed on two-dimensional Gauss-Markov sequences. One-dimensional Hadamard transformation is performed in the horizontal direction and the vertical correlation of the simulated data is exploited to achieve partial error detection and correction. Viewphone signals, particularly those derived from head and shoulder scenes, possess much spatial and temporal redundancy which may be significantly reduced at the expense of a slight, but subjectively tolerable, degree of degregation in picture quality. The discrete Cosine transform offers a corriparatively fast and efficient transformation for achieving this objective, and is used extensively here, in conjunction with various adaptive algorithms which select and code the transform coefficients for transmission on the basis of the statistical and local content of the data. Initially, twodimensional adaptive techniques are employed followed by an extension of the ideas developed to the time domain. It is demonstrated that a marked reduction in the signal transmission rate can be achieved by this method.