Optimal subband decoding

This paper addresses the question of optimal subband decoding, given a system of subband analysis filters and a predetermined quantization strategy. Optimal analysis-synthesis pairs are of great interest, but have been characterized only under very strong (and impractical) conditions. Limiting our attention to the decoder, we find optimal synthesis filters by solving a system of regression equations. Neither high-rate quantization, whiteness of the quantization noise, nor optimality of the quantizer is assumed. Optimal coefficients are independent of the internal machinery of the quantizers, and can therefore be used with "canned" software packages. Only knowledge of the analysis filter bank is assumed. Preliminary experiments show modest gains, on the order of 0.1-0.3 dB, by applying this method to one-stage reproduction of natural images from zerotree quantizers.

在给定子带分析过滤器系统和预定的量化策略的情况下，本文解决了最佳子带解码的问题。最佳分析 - 合成对具有引起极大的兴趣，但仅在非常强（和不切实际的）条件下才进行表征。将注意力限制在解码器上，我们通过求解回归方程系统找到最佳的合成过滤器。既不假定高速量化，量化噪声的白度也不是假定的量化器。最佳系数独立于量化器的内部机械，因此可以与“罐装”软件包一起使用。仅假定分析的知识。初步实验表明，通过将此方法应用于Zerotree量化器的自然图像的一阶段再现，以0.1-0.3 dB的顺序显示了适度的增长。