Volumetric medical image compression.

摘要

This thesis investigates a more efficient volumetric medical image compression algorithm than all other algorithms by 3-D subband/wavelet and presents a three-dimensional (3-D) embedded volumetric coding algorithm that utilizes that 3-D wavelet property and 3-D Set Partitioning in Hierarchical Trees (3-D SPIHT). In order to achieve the embeddedness, we introduce an idea of transaxial-axial orientation tree. By this means, the significance information of 3-D wavelet transformed volumetric sequence is encoded to reduce the Mean-Squared-Error (MSE) distortion measure. By successively partitioning a set of coefficients into its subsets and transmitting significant information first, the algorithm gradually approximates the original volumetric sequence. We demonstrate progressive lossy to lossless compression of volumetric medical images using 3-D integer wavelet packet transforms and 3-D SPIHT. The coding efficiency of the SPIHT algorithm comes from exploiting the self-similarity present in the wavelet multiresolution represent action.; We implement to provide good rate versus distortion performance using low memory and to suppress quality losses at boundaries between the coding units. We consider the division of the image volume uniformly into horizontal stripe sequences with a smaller number of rows and compressing each stripe independently, using the SPIHT algorithm. Our low memory implementation smooths out considerably the variation in mean squared error among different slices and suffers only an insignificant loss in performance from that of a full memory implementation, due mostly to the redundancy introduced by overlapping. This research also investigates the algorithm which deliver higher rate lossy compression within the Region of Interest (ROI), and lower rate lossy compression elsewhere in the image, might be the key to providing efficient and accurate image coding to the medical community.; Additionally, we investigate the line-based transform to reduce memory requirements. We do not need the whole sequence to form the coefficients belonging to a given part of the 3-D transform. So our memory requirement depends on the filter length and not the sequence length. Our results demonstrate that our coder is still very competitive with respect to similar state of the art coders, such as 3-D SPIHT coding.