Nonuniform weighted subsampling for digital image compression

In this paper, weighted subsampling (WS) a new, nonuniform image subsampling scheme is presented. The autoregressive model with one parameter (AR(1)) is used to model the input image. The proposed technique uses a weighting function to change the sampling pattern so that more samples are taken in important regions of the image. The signal is weighted to change its autocorrelation matrix and a sampling pattern is deduced and applied to the image. Predictive lossless coding is used to extend the compression achieved by the subsampling operation. Experiments with facial images demonstrate the effectiveness of the proposed method