A higher order Hopfield network for vector quantisation

A higher order version of the Hopfield neural network is presented which will perform a simple vector quantisation or clustering function. This model requires no penalty terms to impose constraints in the Hopfield energy, in contrast to the usual one where the energy involves only terms quadratic in the state vector. The energy function is shown to have no local minima within the unit hypercube of the state vector so the network only converges to valid final states. Optimisation trials show that the network can consistently find optimal clusterings for small, trial problems and near optimal ones for a large data set consisting of the intensity values from the digitised, grey-level image.

Reduced-memory multirate vector quantisation

A new method is proposed which reduces the size of the memory needed to implement multirate vector quantizers. Investigations have shown that the performance of the coders implemented using this approach is comparable to that obtained from standard systems. The proposed method can therefore be used to reduce the hardware required to implement real-time speech coders.

Systolic VLSI compiler (SVC) for high performance vector quantisation chips

An overview is given of a systolic VLSI compiler (SVC) tool currently under development for the automated design of high performance digital signal processing (DSP) chips. Attention is focused on the design of systolic vector quantization chips for use in both speech and image coding systems. The software in question consists of a cell library, silicon assemblers, simulators, test pattern generators, and a specially designed graphics shell interface which makes it expandable and user friendly. It allows very high performance digital coding systems to be rapidly designed in VLSI.

Variation of vector quantisation and speech waveform coding

A new idea for waveform coding using vector quantisation (VQ) is introduced. This idea makes it possible to deal with codevectors much larger than before for a fixed bit per sample rate. Also a solution to the matching problem (inherent in the present context) in the &-norm describing a measure of neamess is presented. The overall computational complexity of this solution is O(n3 log, n). Sample results are presented to demonstrate the advantage of using this technique in the context of coding of speech waveforms.

Vector quantization based Gaussian modeling for speaker verification

Gaussian mixture models (GMMs) have become an established means of modeling feature distributions in speaker recognition systems. It is useful for experimentation and practical implementation purposes to develop and test these models in an efficient manner particularly when computational resources are limited. A method of combining vector quantization (VQ) with single multi-dimensional Gaussians is proposed to rapidly generate a robust model approximation to the Gaussian mixture model. A fast method of testing these systems is also proposed and implemented. Results on the NIST 1996 Speaker Recognition Database suggest comparable and in some cases an improved verification performance to the traditional GMM based analysis scheme. In addition, previous research for the task of speaker identification indicated a similar system perfomance between the VQ Gaussian based technique and GMMs

Linear filtering methods for fixed rate quantisation with noisy symmetric error channels

This study considers linear filtering methods for minimising the end-to-end average distortion of a fixed-rate source quantisation system. For the source encoder, both scalar and vector quantisation are considered. The codebook index output by the encoder is sent over a noisy discrete memoryless channel whose statistics could be unknown at the transmitter. At the receiver, the code vector corresponding to the received index is passed through a linear receive filter, whose output is an estimate of the source instantiation. Under this setup, an approximate expression for the average weighted mean-square error (WMSE) between the source instantiation and the reconstructed vector at the receiver is derived using high-resolution quantisation theory. Also, a closed-form expression for the linear receive filter that minimises the approximate average WMSE is derived. The generality of framework developed is further demonstrated by theoretically analysing the performance of other adaptation techniques that can be employed when the channel statistics are available at the transmitter also, such as joint transmit-receive linear filtering and codebook scaling. Monte Carlo simulation results validate the theoretical expressions, and illustrate the improvement in the average distortion that can be obtained using linear filtering techniques.

Applications of fractals to image data compression

Digital image processing is exploited in many diverse applications but the size of digital images places excessive demands on current storage and transmission technology. Image data compression is required to permit further use of digital image processing. Conventional image compression techniques based on statistical analysis have reached a saturation level so it is necessary to explore more radical methods. This thesis is concerned with novel methods, based on the use of fractals, for achieving significant compression of image data within reasonable processing time without introducing excessive distortion. Images are modelled as fractal data and this model is exploited directly by compression schemes. The validity of this is demonstrated by showing that the fractal complexity measure of fractal dimension is an excellent predictor of image compressibility. A method of fractal waveform coding is developed which has low computational demands and performs better than conventional waveform coding methods such as PCM and DPCM. Fractal techniques based on the use of space-filling curves are developed as a mechanism for hierarchical application of conventional techniques. Two particular applications are highlighted: the re-ordering of data during image scanning and the mapping of multi-dimensional data to one dimension. It is shown that there are many possible space-filling curves which may be used to scan images and that selection of an optimum curve leads to significantly improved data compression. The multi-dimensional mapping property of space-filling curves is used to speed up substantially the lookup process in vector quantisation. Iterated function systems are compared with vector quantisers and the computational complexity or iterated function system encoding is also reduced by using the efficient matching algcnithms identified for vector quantisers.

IMPROVED POSTFILTERING TECHNIQUE FOR BLOCK QUANTISATION OF SPEECH.

A modified comb filtering technique is proposed which can be used to reduce framing noise generated when speech signals are transform-coded or vector-quantized. Application of this filter to 9. 6 kbit/s speech in a vector transform coder has been found to improve the perceptual quality of the coded speech.