Weighted universal bit allocation: optimal multiple quantization matrix coding

local variation in data statistics. A code that incoruorates a variety of bit allocation options and allows the bitallocawe introduce a two-stage bit allocation algorithm gous to the algorithm for weighted vector quanpossible bit allocations (typically in the form of a collection of quantization matrices) rather than a single bit allocation tion to change from image to image or even from data block to data block can better track local The necessary to describe the chosen bit allocation. (wuvQ) [l, 21. The encoder a collection Of expense of tracking this information is the side information We here present a technique for designing an optimal (Or sing1e quantization matrix). we describe both an encoding algorithm for achieving Optimal compression using collection of bit allocations for a given source tion. The design algorithm is functionally equivalent to the a collection of bit allocations and a technique for designing locally optimal collections of bit allocations. We demonstrate performance on a JPEG style coder using the mean squared error (mse) distortion measure. On a sequence of medical brain scans, the algorithm achieves up to 2.5 dB imwuvQ design strategy eralized Lloyd universal bit allocation (WUBA). cation and the WUBA design algorithm; in Section a], which in turn mirrors the genwe call our technique weighted In Section 2 we describe weighted universal bit allowe PIovement Over a sing1e bit allocation system, up to dB improvement Over a wuvQ with firstand second-stage describe a simple transform code, and in Section 4 we use this code to demonstrate the of WUBA Over a single vector dimensions equal to 16 and 4 respectively, and up to 12 dB imurovement over an entrow constrained vector bit allocation system.