Model-based Vector Quantization Research Articles

Line spectral frequencies modeling by a mixture of von Mises–Fisher distributions

Efficient quantization of the linear predictive coding (LPC) parameters plays a key role in parametric speech coding. The line spectral frequency (LSF) representation of the LPC parameters has found its applications in speech model quantization. In practical implementations of vector quantization (VQ), probability density function optimized VQ has been shown to be more efficient than the VQ based on training data. In this paper, we present the LSF parameters by a unit vector form, which has directional characteristics. The underlying distribution of this unit vector variable is modeled by a von Mises–Fisher mixture model (VMM). An optimal inter-component bit allocation strategy is proposed based on high rate theory and a distortion-rate (D-R) relation is derived for the VMM based-VQ (VVQ). Experimental results show that the VVQ outperforms the recently introduced Dirichlet mixture model-based VQ and the conventional Gaussian mixture model-based VQ, in terms of modeling performance and D-R relation.

An adaptive technique for accuracy enhancement of vector quantizers in nonorthogonal domains

Recently, novel vector quantization techniques in multiple nonorthogonal domains for both waveform and model-based signal characterization that give an improved qualitative and quantitative signal coding performance as compared to vector quantization in single domain have been reported. In these techniques, vectors are formed either directly from the signal waveform or from the model parameters extracted from the signal. Then, the vectors are represented in multiple nonorthogonal domains. The encoder chooses the domain that best represents the vector according to a predetermined criterion. An iterative codebook enhancement algorithm, applicable to both waveform and model-based vector quantization in nonorthogonal domains is developed and presented in this paper. In this algorithm, each set of codebooks in a given domain is retrained by the vectors that were best represented by that particular set of codebooks in the most recent iteration. The algorithm is applied successfully and extensive simulation results yield considerable performance enhancement of the vector quantization in nonorthogonal domains, for a given bit rate. Sample results are provided which demonstrate the improved performance.

Model-based vector quantization with application to remotely sensed image data

Model-based vector quantization (MVQ) is introduced here as a variant of vector quantization (VQ). MVQ has the asymmetrical computational properties of conventional VQ, but does not require the use of pregenerated codebooks. This is a great advantage, since codebook generation is usually a computationally intensive process, and maintenance of codebooks for coding and decoding can pose difficulties. MVQ uses a simple mathematical model for mean removed errors combined with a human visual system model to generate parameterized codebooks. The error model parameter (lambda) is included with the compressed image as side information from which the same codebook is regenerated for decoding. As far as the user is concerned, MVQ is a codebookless VQ variant. After a brief introduction, the problems associated with codebook generation and maintenance are discussed. We then give a description of the MVQ algorithm, followed by an evaluation of the performance of MVQ on remotely sensed image data sets from NASA sources. The results obtained with MVQ are compared with other VQ techniques and JPEG/DCT. Finally, we demonstrate the performance of MVQ as a part of a progressive compression system suitable for use in an image archival and distribution installation.

A Rate-Distortion Function for Vector Quantization with a Variable Block-Size Classification Model

In this paper, a rate-distortion function (RDF),R(D), is presented for a variable block-size classification (VBSC) model. We obtain a theoreticalR(D) bound on the performance of vector quantization (VQ) based on the VBSC model. It is theoretically proved that theR(D) bound of the VBSC model is lower than those of the Gaussian model and the fixed block-size classification (FBSC) model for the bit rates of interest. In the comparison tests of VBSC model-based VQ and FBSC model-based VQ, which were carried out by using a monochrome still image, it was seen that the former technique outperforms the latter technique, subjectively as well as objectively. We also experimentally evaluate a RDF for the VBSC model and compare this with the theoretical RDF. There is a gap of 0.07–0.1 bpp between the theoretical RDF and the experimental RDF in VQ coding without entropy coding. We have reduced the gap to 0.02–0.03 bpp by subsequently employing a Huffman coder for entropy coding. It is expected that the theoretical bound can be approached more closely by the experimental RDF by using a modified asymptotic RDF.