An adaptive two‐stage image coding scheme using dpcm and vector quantization

Abstract

AbstractThis paper proposes a two‐stage coding scheme, whereby a reduced image for retrieval is sent at the first stage, as a highly efficient coding scheme for images and other databases. In the first stage, the reduced image is encoded by block adaptive DPCM. In the second stage, the interpolative error component is encoded using adaptive vector quantization. In this scheme the number of quantization levels is easily determined from the standard deviation of the prediction error signals calculated for each subregion and the rate‐distortion characteristic of the scalar or vector quantizers, using the same method for either of the two stages. By designing the quantizers for typical probability distributions of the predictive error signal, a highly versatile coding scheme is realized.The proposed method has a better overall coding performance than a vector quantizer with the same block size (4 × 4 pixels), which has been designed in an optimal way in the original signal space. Compared with adaptive discrete cosine transform vector quantization [16] and adaptive predictive error vector quantization of the prediction error signals [7], which are also highly versatile, the proposed method exhibits a better characteristic when the coding rate exceeds a certain level. There is little degradation due to the two‐stage coding process. It has also been verified that the proposed method is effective for coding color images.