Distortion estimates for adaptive lifting transforms with noise

Multimedia analysis, enhancement and coding methods often resort to adaptive transforms that exploit local characteristics of the input source. Following the signal decomposition stage, the produced transform coefficients and the adaptive transform parameters can be subject to quantization and/or data corruption (e.g. due to transmission or storage limitations). As a result, mismatches between the analysis- and synthesis-side transform coefficients and adaptive parameters may occur, severely impacting the reconstructed signal and therefore affecting the quality of the subsequent analysis, processing and display task. Hence, a thorough understanding of the quality degradation ensuing from such mismatches is essential for multimedia applications that rely on adaptive signal decompositions. This paper focuses on lifting-based adaptive transforms that represent a broad class of adaptive decompositions. By viewing the mismatches in the transform coefficients and the adaptive parameters as perturbations in the synthesis system, we derive analytic expressions for the expected reconstruction distortion. Our theoretical results are experimentally assessed using 1D adaptive decompositions and motion-adaptive temporal decompositions of video signals.

Figure optionsDownload high-quality image (170 K)Download as PowerPoint slideHighlights
► Our distortion estimates account for noise in the transform coefficients and in the adaptive parameters.
► Starting with the 1D case, we extend our approach to motion-adaptive temporal lifting synthesis of video.
► The analytical estimate matches the experimentally measured average distortion closely.
► The results can be applied to noisy synthesis of any adaptive lifting scheme.