Structured trees for lossless coding of quantized wavelet coefficients

We describe two low-complexity entropy coding structures for compression of quantized wavelet coefficients for image compression applications. The coders operate independently within each subband and represent zero runs and levels. The first coder uses a parameterized, structured coding tree in which the parameters are selected on an image- or subband-specific basis. The second coder uses a set of non-adaptive, non-structured coding trees. In both of the methods the coefficients are considered in raster order on an intrasubband basis. This leads to a simpler structure relative to zerotree algorithms, though in contrast with zerotree, the technique we use does not produce an embedded bitstream. The main advantage lies in the ability of this approach to provide good performance (typically within 1 dB of the best PSNR results reported in the literature) at a cost which is significantly lower than other algorithms providing comparable PSNR results.