Incremental Refinement of Computation for the Discrete Wavelet Transform

摘要: Contrary to the conventional paradigm of transform decomposition followed by quantization, we investigate computation two-dimensional (2D) discrete wavelet transforms (DWT) under quantized representations input source. The proposed method builds upon previous research on approximate signal processing and revisits concept incremental refinement computation: Under a source description (with use an embedded quantizer), forward inverse refines previously computed result, thereby leading output. In first part this paper, study both DWT state-of-the-art 2D lifting-based formulations. By focusing bitplane-based (double-deadzone) propose schemes that achieve for multilevel or reconstruction based bitplane-by-bitplane calculation approach. second part, stochastic modeling typical coefficients, derive analytical model estimate arithmetic complexity computation. is parameterized with respect ( i) operational settings, such as total number levels terminating bitplane; (ii) algorithm-related e.g., variance, related choice wavelet, etc. Based derived formulations, which subsets these parameters framework derives identical accuracy approach without any incurring computational overhead. This termed successive computation, since all representation accuracies are produced incrementally single (continuous) refined no overhead in comparison specifically targets each level not refinable. Our results, well estimates refinement, validated real video sequences compressed scalable coder.