Image coding with vector quantization (VQ) reveals several defects which include edge degradation and high encoding complexity. This paper presents an edge-preserving coding system based on VQ to overcome these defects. A signal processing unit first classifies image blocks into low-activity or high-activity class. A high-activity block is then decomposed into a smoothing factor, a bit-plane and a smoother (lower variance) block. These outputs can be more efficiently encoded by VQ with lower distortion. A set of visual patterns is used to encode the bit-planes by binary vector quantization. We also develop a modified search-order coding to further reduce the redundancy of quantization indexes. Simulation results show that the proposed algorithm achieves much better perceptual quality with higher compression ratio and significant lower computational complexity, as compared to the direct VQ.

This paper presents two competitive learning methods with the objective of avoiding the initial dependency of weight (reference) vectors. The first is termed the refractory and competitive learning algorithm. The algorithm has a refractory period: Once the cell has fired, a winner unit corresponding to the cell is not selected until a certain amount of time has passed. Thus, a specific unit does not become a winner in the early stage of processing. The second is termed the creative and competitive learning algorithm. The algorithm is presented as follows: First, only one output unit is prepared at the initial stage, and a weight vector according to the unit is updated under the competitive learning. Next, output units are created sequentially to a prespecified number based on the criterion of the partition error, and competitive learning is carried out until the ternimation condition is satisfied. Finally, we discuss algorithms which have little dependence on the initial values and compare them with the proposed algorithms. Experimental results are presented in order to show that the proposed methods are effective in the case of average distortion.

An analog vector quantization processor has been designed based on the neuron-MOS (νMOS) technology. In order to achieve a high integrating density, template information is merged into the matching cell (the absolute value circuitry) using the νMOS ROM technology. A new-architecture νMOS winner-take-all (WTA) circuit is employed for fully-parallel search for the minimum-distance vector. The WTA performs multi-resolution winner search with an automatic feedback gain control. A test chip having 256 16-element fixed template vectors has been built in a 1.5-µm double-polysilicon CMOS technology with the chip size of 7.2 mm 7.2 mm, and the basic operation of the circuits has been demonstrated.

Current computing systems are too slow for information processing because of the huge number of procedural steps required. A decrease in the number of calculation steps is essential for real-time information processing. We have developed two kinds of novel architectures for automatic elimination of redundant calculation steps. The first architecture employs the new digit-serial algorithm which eliminates redundant lower digit calculations according to the most-significant-digit-first (MSD-first) digit-serial calculation scheme. Basic components based on this architecture, which employ the redundant number system to limit carry propagation, have been developed. The MSD-first sequential vector quantization processor (VQP) is 3.7 times faster than ordinary digital systems as the result of eliminating redundant lower-bit calculation. The second architecture realizes a decrease in the number of complex calculation steps by excluding useless data before executing the complex calculations according to the characterized value of the data. About 90% of Manhattan-distance (MD) calculations in VQP are excluded by estimating the MD from the average distance.

In this paper, we consider the Integrate-and-Fire Model (ab. IFM) with two periodic inputs. The IFM outputs a pulse-train which is governed by a one dimensional return map. Using the return map, the relationship between the inputs and the output is clarified: the first input determines the global shape of the return map and the IFM outputs various periodic and chaotic pulse-trains; the second input quantizes the state of the return map and the IFM outputs various periodic pulse-trains. Using a computer aided analysis method, the quantized return map can be analyzed rigorously. Also, some typical phenomena are confirmed in the laboratory.

In this paper, a novel variable-rate vector quantizer (VQ) design algorithm using fuzzy clustering technique is presented. The algorithm, termed fuzzy entropy-constrained VQ (FECVQ) design algorithm, has a better rate-distortion performance than that of the usual entropy-constrained VQ (ECVQ) algorithm for variable-rate VQ design. When performing the fuzzy clustering, the FECVQ algorithm considers both the usual squared-distance measure, and the length of channel index associated with each codeword so that the average rate of the VQ can be controlled. In addition, the membership function for achieving the optimal clustering for the design of FECVQ are derived. Simulation results demonstrate that the FECVQ can be an effective alternative for the design of variable-rate VQs.

We propose a motion estimation algorithm using less gray level images, which are composed of bits pixels lower than 8 bits pixels. Threshold values for generating low bits pixels from 8 bits pixels are simply determined as median values of pixels in a macro block. The proposed algorithm reduces the computational complexity of motion estimation at less expense of video quality. Moreover, median cut quantization can be applied to multilevel images and combined with a lot of fast algorithms to obtain more effective algorithms.

In this paper, the generalized learning vector quantization (GLVQ) algorithm is applied to design a hand-written Chinese character recognition system. The system proposed herein consists of two modules, feature transformation and recognizer. The feature transformation module is designed to extract discriminative features to enhance the recognition performance. The initial feature transformation matrix is obtained by using Fisher's linear discriminant (FLD) function. A template matching with minimum distance criterion recognizer is used and each character is represented by one reference template. These reference templates and the elements of the feature transformation matrix are trained by using the generalized learning vector quantization algorithm. In the experiments, 540100 (5401 100) hand-written Chinese character samples are used to build the recognition system and the other 540100 (5401 100) samples are used to do the open test. A good performance of 92.18 % accuracy is achieved by proposed system.

In this paper, a classified vector quantization (CVQ) method using a novel direction based classifier is proposed. The new classifier uses a distortion measure related to the angle between vectors to determine the similarity of vectors. The distortion measure is simple and adequate to classify various edge types other than single and straight line types, which limit the size of image block to a rather small size. Simulation results show that the proposed technique can achieve better perceptual quality and edge integrity at a larger block size, as compared to other CVQs. It is shown when the vector dimension is changed from 16(4 4) to 64(8 8), the average bit rate can be reduced from 0. 684 bpp to 0.191, whereas the PSNR degradation is only about 1.2 dB.

We propose a real-time low-rate video compression algorithm using fixed-rate multi-stage hierarchical vector quantization. Vector quantization is suitable for mobile computing, since it demands small computation on decoding. The proposed algorithm enables transmission of 10 QCIF frames per second over a low-rate 29.2 kbps mobile channel. A frame is hierarchically divided by sub-blocks. A frame of images is compressed in a fixed rate at any video activity. For active frames, large sub-blocks for low resolution are mainly transmitted. For inactive frames, smaller sub-blocks for high resolution can be transmitted successively after a motion-compensated frame. We develop a compression system which consists of a host computer and a memory-based processor for the nearest neighbor search on VQ. Our algorithm guarantees real-time decoding on a poor CPU.

We introduce an image contour clustering method based on a multiscale image representation and its application to image compression. Multiscale gradient planes are obtained from the mean squared sum of 2D wavelet transform of an image. The decay on the multiscale gradient planes across scales depends on the Lipshitz exponent. Since the Lipshitz exponent indicates the spatial differentiability of an image, the multiscale gradient planes represent smoothness or sharpness around edges on image contours. We apply vector quatization to the multiscale gradient planes at contours, and cluster the contours in terms of represntative vectors in VQ. Since the multiscale gradient planes indicate the Lipshitz exponents, the image contours are clustered according to its gradients and Lipshitz exponents. Moreover, we present an image recovery algorithm to the multiscale gradient planes, and we achieve the skech-based image compression by the vector quantization on the multiscale gradient planes.

Macroscopic method for quantization of the evanescent fields brought about by total reflection is presented. Here, a semi-infinite space is assumed to be filled with a transparent dispersive dielectric with dielectric constant ε(ω) to the left of the plane z = 0, and be empty to the right of the plane. The wave is assumed to be incident from the left, and so the whole field is composed of the triplet of incident, reflected, and transmitted waves labeled by a continuous wave vector index. The transmitted wave in free space may be evanescent. The triplet is shown exactly without using slowly varying field approximation in dispersive medium to form orthogonal mode for different wave vectors, which provides the basis for the quantization of the triplet with taken into account of medium dispersion. The exact orthogonal relation reduces to the well known one if the dielectric is nondispersive, ε/ω = 0. By using the field expansion in terms of the orthogonal triplet modes, the total field energy is found to be the sum of the energies of independent harmonic oscillators. A discussion is also made on the wave momentum of evanescent field.

To improve speech coding quality, in particular, the long-term dependency prediction characteristics, we propose a new nonlinear predictor, i. e. , a fully connected recurrent neural network (FCRNN) where the hidden units have feedbacks not only from themselves but also from the output unit. The comparison of the capabilities of the FCRNN with conventional predictors shows that the former has less prediction error than the latter. We apply this FCRNN instead of the previously proposed recurrent neural networks in the code-excited predictive speech coding system (i. e. , CELP) and shows that our system (FCRNN) requires less bit rate/frame and improves the performance for speech coding.

In this paper, the application of an unsupervised parallel approach called the Fuzzy Hopfield Neural Network (FHNN) for vector qunatization in image compression is proposed. The main purpose is to embed fuzzy reasoning strategy into neural networks so that on-line learning and parallel implementation for codebook design are feasible. The object is to cast a clustering problem as a minimization process where the criterion for the optimum vector qunatization is chosen as the minimization of the average distortion between training vectors. In order to generate feasible results, a fuzzy reasoning strategy is included in the Hopfield neural network to eliminate the need of finding weighting factors in the energy function that is formulated and based on a basic concept commonly used in pattern classification, called the "within-class scatter matrix" principle. The suggested fuzzy reasoning strategy has been proven to allow the network to learn more effectively than the conventional Hopfield neural network. The FHNN based on the within-class scatter matrix shows the promising results in comparison with the c-means and fuzzy c-means algorithms.

This paper presents a novel classified vector quantizer (CVQ) design algorithm which can control the rate and storage size for applications of image coding. In the algorithm, the classification of image blocks is based on the edge orientation of each block in the wavelet domain. The algorithm allocates the rate and storage size available to each class of the CVQ optimally so that the average distortion is minimized. To reduce the arithmetic complexity of the CVQ, we employ a partial distance codeword search algorithm in the wavelet domain. Simulation results show that the CVQ enjoys low average distortion, low encoding complexity, high visual perception quality, and is well-suited for very low bit rate image coding.

This letter presents a novel variable-rate vector quantizer (VQ) design algorithm, which is a hybrid approach combining a genetic algorithm with the entropy-constrained VQ (ECVQ) algorithm. The proposed technique outperforms the ECVQ algorithm in the sense that it reaches to a nearby global optimum rather than a local one. Simulation results show that, when applied to the image coding, the technique achieves higher PSNR and image quality than those of ECVQ algorithm.

We present a mechanism, named the law of the jungle (LOJ), to improve the Kohonen learning. The LOJ is used to be an adaptive vector quantizer for approximating nonstationary probability distribution functions. In the LOJ mechanism, the probability that each node wins in a competition is dynamically estimated during the learning. By using the estimated win probability, "strong" nodes are increased through creating new nodes near the nodes, and "weak" nodes are decreased through deleting themselves. A pair of creation and deletion is treated as an atomic operation. Therefore, the nodes which cannot win the competition are transferred directly from the region where inputs almost never occur to the region where inputs often occur. This direct "jump" of weak nodes provides rapid convergence. Moreover, the LOJ requires neither time-decaying parameters nor a special periodic adaptation. From the above reasons, the LOJ is suitable for quick approximation of nonstationary probability distribution functions. In comparison with some other Kohonen learning networks through experiments, only the LOJ can follow nonstationary probability distributions except for under high-noise environments.

We have developed and fabricated an LSI called the FMPP-VQ64. The LSI is a memory-based shared-bus SIMD parallel processor containing 64 PEs, intended for low bit-rate image compression using vector quantization. It accelerates the nearest neighbor search (NNS) during vector quantization. The computation time does not depend on the number of code vectors. The FMPP-VQ64 performs 53,000 NNSs per second, while its power dissipation is 20 mW. It can be applied to the mobile telecommunication system.

This paper focuses on competitive learning algorithms for WTA (winner-take-all) networks which perform rotation invariant pattern classification. Although WTA networks may theoretically be possible to achieve rotation invariant pattern classification with infinite memory capacities, actual networks cannot memorize all input data. To effectively memorize input patterns or the vectors to be classified, we present two algorithms for learning vectors in classes (LVC1 and LVC2), where the cells in the network memorize not only weight vectors but also their firing numbers as statistical values of the vectors. The LVC1 algorithm uses simple and ordinary competitive learning functions, but it incorporates the firing number into a coefficient of the weight change equation. In addition to all the functions of the LVC1, the LVC2 algorithm has a function to utilize under-utilized weight vectors. From theoretical analysis, the LVC2 algorithm works to minimize the energy of all weight vectors to form an effective memory. From computer simulation with two-dimensional rotated patterns, the LVC2 is shown to be better than the LVC1 in learning and generalization abilities, and both are better than the conventional Kohonen self-organizing feature map (SOFM) and the learning vector quantization (LVQ1). Furthermore, the incorporation of the firing number into the weight change equation is shown to be efficient for both the LVC1 and the LVC2 to achieve higher learning and generalization abilities. The theoretical analysis given here is not only for rotation invariant pattern classification, but it is also applicable to other WTA networks for learning vector quantization.

This paper proposes an analysis method of scaling-factor-quantization error in fractal image coding using a state-space approach with the statistical analysis method. It is shown that the statistical analysis method is appropriate and leads to a simple result, whereas the deterministic analysis method is not appropriate and leads to a complex result for the analysis of fractal image coding. We derive the output error variance matrix for the measure of error and define the output error variance by scalar quantity as the mean of diagonal elements of the output error variance matrix. Examples are given to show that the scaling-factor-quantization error due to iterative computation with finite-wordlength scaling factors degrades the quality of decoded images. A quantitative comparison of experimental scaling-factor-quantization error with analytical result is made for the output error variance. The result shows that our analysis method is valid for the fractal image coding.