Sub-band coding of images with quadtree-guided recursive polynomial decomposition

The authors develop a recursive polynomial transform coder and incorporate it in a 16-band subband coder (SBC). The polynomial coder segments an image into subblocks of different sizes and approximates the pixel intensities in a block with a 2-D polynomial. An orthonormal basis set that spans the space of sampled polynomials is determined, and the coefficients of these basis polynomials are then quantized using entropy-constrained scalar quantizers. The lowest band of the SBC is encoded using the polynomial coder, and the upper bands are encoded using uniform threshold entropy-constrained quantizers. The system is then simulated in software and is shown to have a gain of 1.3 dB in peak-to-peak signal-to-noise ratio (PSNR) at 0.25 bits/pixel over P. Strobach´s recursive plane decomposition coder (IEEE Trans. vol.SP-39, p.1380-97, June 1991).<>