3D medical image coding using separable 3D wavelet decomposition and lattice vector quantization

Abstract

An original 3D subband coding scheme is proposed. A separable 3D wavelet, taking full advantage of the 3D structures correlation, decomposes the original volume into subvolumes which can be separately quantized by an uniform scalar quantizer or by a 3D lattice vector quantizer. Concentric hyper-pyramids lying on the cubic lattice are used for searching codewords. A distortion minimization algorithm both selects the best number of decomposition and the best set of quantizers in order to minimize the overall mean square error. The whole algorithm is applied on a 3D image data base issued from the Morphometer (a new true 3D X-Ray scanner). The results presented include traditional signal to noise ratio performances and a subjective evaluation made by radiologists.

Lossy compression of volumetric medical images with 3D dead-zone lattice vector quantization

“…First, we compare the compression performances of our method with those of both classical 3D LVQ [4] and 3D-SPIHT [10] [11]. All the images have been compressed by the three methods at rates varying from 0.05 to 1 bit/voxel (bpv).…”

“…The most approaches combine a threedimensional space decorrelating transform with the extension of a coding algorithm that has proven to be effective on 2D images [3]. For example, Catin et al [4] were the first, to our knowledge, to compress volumetric medical images with Lattice Vector Quantization (LVQ), while many works had been done before on LVQ [5] [6] in the field of 2D real life images.…”

“…In this paper, we propose the design of a new entropycoded LVQ codebook enclosing a multidimensional dead zone to encode the coefficients of a 3D Dyadic Wavelet Transform (3D DWT) which is known to be one of the most efficient decorrelator for 3D medical images [3] [4].…”

3D Steerable Wavelets in Practice

“…The multiresolution backbone of our method is the tight wavelet frame {ψ i,k } i∈Z,k∈Z 3 , whose basis functions in L 2 (R 3 ) are generated by suitable dilations and translations of a single mother wavelet ψ(x). The normalized wavelets at scale i and translation parameter k are given by…”

“…In particular, wavelet-based techniques have been developed for MRI reconstruction [1], 3D deconvolution microscopy [2], and medical image coding [3]. Various 3D wavelets have also been designed and used for video processing [4]- [7].…”