An upper bound of the Rate-Distortion function (RDF) is obtained for a sample image source. It is decorrelated first by the discrete Fouier transform and a linear prediction. The RDF of the transformed and quantized source data is then obtained by the Blahut's computational algorithm. It is compared with the bounds for several Gaussian covariance models. It is found that the bound is lower than the Gaussian bound by 0.2 bit/pel.
Also this bound is compared with the data rates of some coding algorithms including DPCM/CRC. DPCM/CRC is an extension of the conditional runlength coding (CRC) to general images. In this comparison, we found that the performance of DPCM/CRC is better than those of differential pules coded modulation (DPCM) or discrete cosine transform (DCT).
