Adaptive downsampling to improve image compression at low bit rates

摘要: At low bit rates, better coding quality can be achieved by downsampling the image prior to compression and estimating missing portion after decompression. This paper presents a new algorithm in such paradigm, based on adaptive decision of appropriate directions/ratios quantization steps, order achieve higher with rates consideration local visual significance. The full-resolution restored from DCT coefficients downsampled pixels so that spatial interpolation required otherwise is avoided. proposed significantly raises critical rate approximately 1.2 bpp, 0.15-0.41 bpp existing downsample-prior-to-JPEG schemes and, therefore, outperforms standard JPEG method much wider bit-rate scope. experiments have demonstrated PSNR improvement over techniques before rate. In addition, mode not only makes less image-independent, but also automates switching coders variable applications, since turns whenever it necessary at