Predictive naturalness preserving transform coding of monochrome images

Transform predictive coding of pictorial data using the naturalness-preserving transform (NPT) is investigated. The hybrid coding system uses one-dimensional NPTs and parallel differential pulse-code modulation (DPCM) systems. The problem of determining the basis giving the minimum mean square error (MSE) for coding images is addressed. Results of coding schemes using the Hadamard NPT and the sine NPT are presented. Superior performance was achieved by adapting the NPT bases to image subarray statistics. The MSE improvement is higher at low bit rates.<>