Skip to Main content Skip to Navigation

Contributions en compression d'images médicales 3D et d'images naturelles 2D

Abstract : In this thesis, we propose a new lossy coding scheme based on 3D Wavelet Transform and Dead Zone Lattice Vector Quantization 3D (DZLVQ) for medical images. Our algorithm has been evaluated on several CT an MR image volumes. The main contribution of this work is the design of a multidimensional dead zone which enables to take into account correlations between neighbor voxels. At high compression ratios, we show that it can outperform visually and numerically the best existing methods. These promising results are confirmed on head CT by two medical patriciansThe second contribution of this document assesses the effect lossy image compression on CAD (Computered-Aided Decision) detection performance of solid lung nodules. This work on 120 significant lungs MTCDT shows that detection did not suffer until 48:1 compression and still was robust at 96:1.The last contribution consists in the complexity reduction of our compression scheme. The first allocation dedicated to 2D DZLVQ uses an exponential of the rate-distortion (R-D) functions. The second allocation for 2D and 3D medical images is based on block statistical model to estimate the R-D curves. These R-D models are based on the joint distribution of wavelet vectors using a multidimensional mixture of generalized Gaussian (MMGG) densities.
Document type :
Complete list of metadata
Contributor : Thèses Ul Connect in order to contact the contributor
Submitted on : Thursday, March 29, 2018 - 11:26:38 AM
Last modification on : Saturday, October 16, 2021 - 11:14:14 AM
Long-term archiving on: : Friday, September 14, 2018 - 12:30:08 AM


Files produced by the author(s)


  • HAL Id : tel-01748152, version 1



Yann Gaudeau. Contributions en compression d'images médicales 3D et d'images naturelles 2D. Autre. Université Henri Poincaré - Nancy 1, 2006. Français. ⟨NNT : 2006NAN10158⟩. ⟨tel-01748152⟩



Record views


Files downloads