ISBN 9783843947961

978-3-8439-4796-1, Reihe Ingenieurwissenschaften

Daniela Wokusch
Extended Compensated Wavelet Lifting for Scalable Lossless Coding of Dynamic Medical Data

188 Seiten, Dissertation Universität Erlangen-Nürnberg (2020), Softcover, A5

Zusammenfassung / Abstract

Compensated wavelet lifting describes a promising approach in scalable lossless video coding of dynamic medical data. By encoding the resulting low- and highpass subbands with JPEG 2000, a fully scalable bit stream, providing DICOM compatibility, can be generated. To further improve the data fidelity as well as the coding efficiency, three main extensions of compensated wavelet lifting have been developed within the scope of this thesis.

The first extension describes a novel method for compensating motion within dynamic medical data based on signal processing on graphs. To guarantee lossless reconstruction at the decoder side, several coding steps are proposed for efficient transmission of the required side information. However, it can be observed that for dynamic medical data the incorporation of any motion compensation method performs less efficient in terms of compression than uncompensated wavelet lifting. Therefore, the second extension addresses the enhancement of the compression efficiency of compensated wavelet lifting of dynamic medical data using denoising filters in the prediction and update step of the lifting structure. Consequently, the required rate for encoding the single subbands can be reduced significantly, while the data fidelity of the lowpass subband regarding the original sequence can be kept at a constantly high level. Further rate reductions can be achieved by recursively applying the temporal wavelet transform to the lowpass subband of the previous stage. However, this might be less efficient for scenes with a large dynamic extent of motion due to occurring artifacts and a decrease of the data fidelity of the final lowpass subband. To compensate this drawback, in the third extension an adaption of the temporal decomposition depth to the content of the input data is developed. This aims at a non-uniform subsampling, assigning a lower temporal resolution to nearly static scenes and a high temporal resolution to high dynamic scenes.