Mr. Adrian Munteanu is a colleague of Euromed-partner, Dr. Paul Cristea from the University of Bucharest. At the Free University of Brussels, he is involved in research projects on data compression. Before the ITIS audience, he presented an advanced method of wavelet lossless compression, which outperforms the classical zero-tree coding. Compression algorithms including integer wavelet transforms and the new significance maps coding techniques are suitable for applications that require remote browsing through large image data sets and real time archiving of medical images.
In standard wavelet compression techniques, it is not possible to reconstruct the lossless version of the original image because of the rounding to integers of the wavelet coefficients. Since final medical diagnosis often has to be performed on the original images, the lossless compression schemes are required for clinical use in medical data base management and telediagnosis applications. Mr. Munteanu and his colleagues introduced a lifting scheme in order to compute an integer wavelet representation. It enabled them to build non linear wavelet transforms, mapping integers to integers, thus permitting a lossless representation of the image pixels.
These kind of transforms allow to transmit the medical images in a progressive way, so that first, a transfer of the lower resolution version of the image is being performed, followed by the successive transmission of the refinement details. Both the forward and inverse wavelet transforms display the same computational complexity and operate at a considerably greater speed than the wavelet routines which use standard filters. In the quantisation module, new techniques are being applied for the coding of the coefficient's significance maps. They are more effective than the traditional zero-tree representation, which organises the information in trees.
Experimental results performed on a set of coronary angiographic images have proven that in lossless compression, the proposed square partitioning coding algorithm and the lattice partitioning algorithm have similar performances as the state-of-the-art context adaptive lossless image coder (CALIC), and outperform with 0.38 bpp the embedded zero-tree coder combined with the lifting scheme, with 0.20 bpp the set partitioning coder, and with 0.70 bpp the lossless JPEG. The newly developed algorithms are extremely suitable for compression schemes in which the received information is gradually refined up to the lossless version of the input data. The information in this article is based on the paper "Wavelet Lossless Compression of Coronary Angiographic Images", written by Mr. Munteanu and his colleague-researchers. You can consult it at the Web site for Electronics and Digital Signal Processing of the Free University of Brussels.