Analog data compression of images via predictive transform coding

It is shown that a recently offered minimum MSE linear predictive transform (LPT) coding formulation inherently leads to a simple and highly efficient continuous amplitude (analog) data compression algorithm. In the proposed scheme only the most energetic unquantized coefficient errors produced by the encoder are transmitted to effect the desired analog data compression. The technique is illustrated with NTSC monochrome images having a bandwidth of 3.6 MHz, and a communication channel characterized by additive zero mean white noise. It is demonstrated that when half of the coefficient errors are transmitted, that is the bandwidth of the transmitted coefficient error signal has been reduced from 3.6 MHz to 1.8 MHz, the reconstructed signal is very close to SNR and visual performance to that obtained when all the coefficient errors are transmitted