Narrow Results

Prototype based classifiers are effective algorithms in modeling classification problems and have been applied in multiple domains. While many supervised learning algorithms have been successfully extended to kernels to improve the discrimination power by means of the kernel concept, prototype based classifiers are typically still used with Euclidean distance measures. Kernelized variants of prototype based classifiers are currently too complex to be applied for larger data sets. Here we propose an extension of Kernelized Generalized Learning Vector Quantization (KGLVQ) employing a sparsity and approximation technique to reduce the learning complexity. We provide generalization error bounds and experimental results on real world data, showing that the extended approach is comparable to SVM on different public data.

Implantable high-density multichannel neural recording microsystems provide simultaneous recording of brain activities. Wireless transmission of the entire recorded data causes high bandwidth usage, which is not tolerable for implantable applications. As a result, a hardware-friendly compression module is required to reduce the amount of data before it is transmitted. This paper presents a novel compression approach that utilizes a spike extractor and a vector quantization (VQ)-based spike compressor. In this approach, extracted spikes are vector quantized using an unsupervised learning process providing a high spike compression ratio (CR) of 10–80. A combination of extracting and compressing neural spikes results in a significant data reduction as well as preserving the spike waveshapes. The compression performance of the proposed approach was evaluated under variant conditions. We also developed new architectures such that the hardware blocks of our approach can be implemented more efficiently. The compression module was implemented in a 180-nm standard CMOS process achieving a SNDR of 14.49dB and a classification accuracy (CA) of 99.62% at a CR of 20, while consuming 4$\mu$W power and 0.16mm² chip area per channel.

In this paper, a new Hopfield-model net called Compensated Fuzzy Hopfield Neural Network (CFHNN) is proposed for vector quantization in image compression. In CFHNN, the compensated fuzzy c-means algorithm, modified from penalized fuzzy c-means, is embedded into Hopfield neural network so that the parallel implementation for codebook design is feasible. The vector quantization can be cast as an optimal problem that may also be regarded as a minimization of a criterion defined as a function of the average distortion between training vector and codevector. The CFHNN is trained to classify the divided vectors on a real image into feasible class to generate an available codebook when the defined energy function converges to near global minimum. The training vectors on a divided image are mapped to a two-dimensional Hopfield neural network. Also the compensated fuzzy c-means technique is used to update the quantization performance and to eliminate searching for the weighting factors. In the context of vector quantization, each training vector on the divided image is represented by a neuron which is fully connected by the other neurons. After a number of iterations, neuron states are refined to reach near optimal result when the defined energy function is converged.

In this paper we consider a hidden Markov mesh random field (HMMRF) for character recognition. The model consists of a "hidden" Markov mesh random field (MMRF) and an overlying probabilistic observation function of the MMRF. Just like the 1-D HMM, the hidden layer is characterized by the initial and the transition probability distributions, and the observation layer is defined by distribution functions for vector-quantized (VQ) observations.

The HMMRF-based method consists of two phases: decoding and training. The decoding and the training algorithms are developed using dynamic programming and maximum likelihood estimation methods. To accelerate the computation in both phases, we employed a look-ahead scheme based on maximum marginal it a posteriori probability criterion for third-order HMMRF. Tested on a larget-set handwritten Korean Hangul character database, the model showed a promising result: up to 87.2% recognition rate with 8 state HMMRF and 128 VQ levels.

In this paper, a novel gray-level image-hiding scheme is proposed. The goal of this scheme is to hide multiple important gray-level images into another meaningful gray-level image. The secret images to be protected are first compressed using the vector quantization scheme. Then, the DES cryptosystem is conducted on the VQ indices and related parameters to generate the encrypted message. Finally, the encrypted message is embedded into the rightmost two bits of each pixel in the cover image.

According to the experimental results, average image qualities of 44.320 dB and 30.885 dB are achieved for the embedded images and the retrieved secret images, respectively. In other words, multiple secret images can be effectively hidden into one host image of the same size. In addition, the proposed scheme strengthens the protection of the secret images by conducting the DES cryptosystem on the related parameters and the VQ indices of the compressed secret images. Therefore, the proposed scheme provides a secure approach to embed multiple important images into another meaningful image of the same size.

In this paper, we propose an efficient secret sharing scheme without pixel expansion. The scheme first uses the VQ-compression method to compress a secret image. This allows senders to share a larger secret image than other methods. Moreover, the proposed method also allows participants to reconstruct a lossless secret image. The generated shadows are meaningful with high quality, so the image does not attract any suspicion from attackers. Because the scheme uses XOR operation during the construction and revealing phases, it is suitable for secret sharing applications.

Genetic Algorithm (GA) has been successfully applied to codebook design for vector quantization and its candidate solutions are normally turned by LBG algorithm. In this paper, to solve premature phenomenon and falling into local optimum of GA, a new Genetic Simulated Annealing-based Kernel Vector Quantization (GSAKVQ) is proposed from a different point of view. The simulated annealing (SA) method proposed in this paper can approach the optimal solution faster than the other candidate approaches. In the frame of GA, firstly, a new special crossover operator and a mutation operator are designed for the partition-based code scheme, and then a SA operation is introduced to enlarge the exploration of the proposed algorithm, finally, the Kernel function-based fitness is introduced into GA in order to cluster those datasets with complex distribution. The proposed method has been extensively compared with other algorithms on 17 datasets clustering and four image compression problems. The experimental results show that the algorithm can achieve its superiority in terms of clustering correct rate and peak signal-to-noise ratio (PSNR), and the robustness of algorithm is also very good. In addition, we took “Lena” as an example and added Gaussian noise into the original image then adopted the proposed algorithm to compress the image with noise. Compared to the original image with noise, the reconstructed image is more distinct, and with the parameter value increasing, the value of PSNR decreases.

A long-term research project toward Mandarin speech recognition techniques for very large vocabulary and unlimited text is considered. By carefully examining the special structures of Chinese language, the first-stage goal is set to be the design of efficient techniques to recognize the finals of Mandarin syllables. In this paper, three special approaches to do this are proposed. The Segmental Model Approach defines the final models by dividing the finals into several segments according to the acoustic structures of the speech signals. The Three-pass Approach uses three consecutive passes to classify the finals into small sets and improve the recognition efficiency. The Multi-section Vector Quantization (MSVQ) Approach, on the other hand, significantly reduces the necessary computation time by incorporating the branch-and-bound algorithm and common codebook concept with the MSVQ techniques. Extensive computer simulations are performed first to optimize each approach by choosing the best set of parameters then to compare the performance of the three approaches. It was found that all the three approaches are very efficient in terms of relatively high recognition rate and short computation time, and the MSVQ Approach provides the highest recognition rate at the shortest computation time, thus it is most attractive.

In a long-term research project, the recognition of Mandarin speech for very large vocabulary and unlimited text is considered. Its first stage goal is to recognize the Mandarin syllables. In a previous paper, an initial/final two-phase recognition approach to recognize these very confusing syllables was proposed, in which each syllable is divided into initial and final parts and recognized separately, and efficient recognition techniques for the finals were proposed and discussed. This paper serves as a continuation and proposes an efficient system to recognize the Mandarin initials. In this system, a classification procedure is first used to categorize the unknown initials into two groups C₁ and C₂; different approaches are then separately applied and independently optimized to recognize C₁ and C₂. It is found that Finite State Vector Quantization (FSVQ) is very useful, whose two modified versions, Modified FSVQ (MFSVQ) and the Second Order FSVQ (SOFSVQ), can provide the best recognition performance for C₁ and C₂ by carefully adjusting a design parameter called characteristic interval. Experimental results show that a recognition rate of 94.1% to 94.7% can be achieved using this system. Such a design is accomplished by carefully considering the special characteristics of Mandarin syllables and initials.

The output layer of a feedforward neural network approximates nonlinear functions as a linear combination of a fixed set of basis functions, or "features". These features are learned by the hidden-layer units, often by a supervised algorithm such as a back-propagation algorithm. This paper investigates features which are optimal for computing desired output functions from a given distribution of input data, and which must therefore be learned using a mixed supervised and unsupervised algorithm. A definition is proposed for optimal nonlinear features, and a constructive method, which has an iterative implementation, is derived for finding them. The learning algorithm always converges to a global optimum and the resulting network uses two layers to compute the hidden units. The general form of the features is derived for the case of continuous signal input, and this result is related to transmission of information through a bandlimited channel. The results of other algorithms can he compared to the optimal features, which in some cases have easily computed closed-form solutions. The application of this technique to the inverse kinematics problem for a simulated planar two-joint robot arm is demonstrated here.

System identification is the term scientists and engineers use to refer to the process of building mathematical models of dynamical systems based on observed data. This paper approaches system identification as a pattern recognition problem. We use computers to simulate the system response for a variety of different mathematical models. For each distinct system model, simulated system responses tend to remain segregated in one or more amorphous regions of system response space despite (1) large variations in system parameters, (2) experimental errors, and (3) noise. The actual system response is classified with the model corresponding to the region of system response space where it is found. The classifier is an Artificial Neural Network (ANN) which implements a Generalized Vector Quantizer (GVQ). A small number of simple but powerful discriminant functions facilitate the correct classification of most of the responses in any given region. The required distribution of discriminants among the regions evolves automatically as they learn their respective functions.

A novel Discriminative Vector Quantization method for Speaker Identification (DVQSI) is proposed, and its parameters selection is discussed. In the training mode of this approach, the vector space of speech features is divided into a number of regions. Then, a Vector Quantization (VQ) codebook for each speaker in each region is constructed. For every possible combination of speaker pairs, a discriminative weight is assigned for each region, based on the region's ability to discriminate between the speaker pair. Consequently, the region, which contains a larger distribution difference between the speech feature vector sets of the two speakers in the speaker pair, plays a more important role by assigning it a larger discriminative weight, in identifying the better speaker match from the two speakers. In the testing mode, to identify an unknown speaker, discriminative weighted average VQ distortion pairs are computed for the unknown speaker input waveform. Then, a technique is described that figures out the best match between the unknown waveform and speakers' templates. The proposed DVQSI approach can be considered a generalization of the existing VQ technique for Speaker Identification (VQSI). The method presented here yields better Speaker Identification (SI) accuracy by employing the discriminative weights and space segmentation as design parameters. This is confirmed experimentally. In addition, a computationally efficient implementation of the DVQSI technique is given which uses a tree-structured-like approach to obtain the codebooks.

In this paper, we present a new generic architectural approach of a Self-Organizing Map (SOM). The proposed architecture, called the Diagonal-SOM (D-SOM), is described as an Hardware–Description-Language as an intellectual property kernel with easily adjustable parameters.The D-SOM architecture is based on a generic formalism that exploits two levels of the nested parallelism of neurons and connections. This solution is therefore considered as a system based on the cooperation of a distributed set of independent computations. The organization and structure of these calculations process an oriented data flow in order to find a better treatment distribution between different neuroprocessors. To validate the D-SOM architecture, we evaluate the performance of several SOM network architectures after their integration on a Xilinx Virtex-7 Field Programmable Gate Array support. The proposed solution allows the easy adaptation of learning to a large number of SOM topologies without any considerable design effort. $16\times 16$ SOM hardware is validated through FPGA implementation, where temporal performance is almost twice as fast as that obtained in the recent literature. The suggested D-SOM architecture is also validated through simulation on variable-sized SOM networks applied to color vector quantization.

The self-organizing map (SOM) is a popular neural network which was designed for solving problems that involve tasks such as clustering and visualization. Especially, it provides a new strategy of clustering using a competition and co-operation principal. The probabilistic Kohonen network (PRSOM) is the stochastic version of classical one. However, determination of the optimal number of neurons, their initial weights vector and their deviation matrix is still a big problem in the literature. These parameters have a great impact on the learning process of the network, the convergence and the quality of results. Also determination of clusters’ number is a very difficult task. In this paper we propose a new method, called H-PRSOM, which looks for the optimal architecture of the map and determines the suitable codebook for speech compression. According to his hierarchical process, H-PRSOM identifies automatically, in each iteration, new initial parameters of the map. The generated parameters will be used in the learning phase of the probabilistic network. Due to its important propriety of initialization and optimization, we expect that the use of this new version of PRSOM algorithm in the vector quantization might provide good results. In order to evaluate its performance, H-PRSOM model is applied to the problem of speech compression of Arabic digits. The conducted experiments show that the proposed method is able to realize the expected goals.

In this paper, we propose a speech recognition algorithm which utilizes hidden Markov models (HMM) and Viterbi algorithm for segmenting the input speech sequence, such that the variable-dimensional speech signal is converted into a fixed-dimensional speech signal, called TN vector. We then use the fuzzy perceptron to generate hyperplanes which separate patterns of each class from the others. The proposed speech recognition algorithm is easy for speaker adaptation when the idea of "supporting pattern" is used. The supporting patterns are those patterns closest to the hyperplane. When a recognition error occurs, we include all the TN vectors of the input speech sequence with respect to the segmentations of all HMM models as the supporting patterns. The supporting patterns are then used by the fuzzy perceptron to tune the hyperplane that can cause correct recognition, and also tune the hyperplane that resulted in wrong recognition. Since only two hyperplane need to be tuned for a recognition error, the proposed adaptation scheme is time-economic and suitable for on-line adaptation. Although the adaptation scheme cannot ensure to correct the wrong recognition right after adaptation, the hyperplanes are tuned in the direction for correct recognition iteratively and the speed of adaptation can be adjusted by a "belief" parameter set by the user. Several examples are used to show the performance of the proposed speech recognition algorithm and the speaker adaptation scheme.

In this paper, we investigate a novel method for an individual's handwritten Chinese character font generation, using stroke correspondence between the reference character database and the compressed character database, by vector quantization. Chinese characters are composed of a combination of radicals. A radical may be separated into several strokes, with each stroke corresponding to two or more common strokes. By paying attention to the characteristics of Chinese characters and the strokes that form them, we consider each stroke to be a vector and compress the stroke pattern using vector quantization. A compression rate of 1.27% is achieved by the vector quantization. We performed the evaluation experiments using both subjective and objective criteria involving 26 subjects and demonstrated that fonts generated successfully reflect the user's individual handwriting.

Three image compression schemes based on vector quantization are proposed in this paper. The block similarity property among neighboring image blocks is exploited in these schemes to cut down the bit rate of the vector quantization scheme. For the first scheme, the correlation among the encoded block to the left and the encoded block directly above the current processing block is exploited. In the second scheme, the relative addressing technique is incorporated into the encoding procedure. Finally, the third scheme introduces a simple technique to reduce the required bit rate with only a slight reduction in image quality. According to the experimental results, it is shown that these proposed schemes not only reduce storage costs but also achieve good reconstructed image quality. Furthermore, the required computational cost for the encoding/decoding procedures of these schemes is less than that of the conventional vector quantization scheme. In other words, these schemes are suitable for the compression of digital images.

In this paper, we present a novel method for the optimal texture sampling and texture synthesis using vector quantization clustering. The synthesizing process selects an initial texture block from the input sample texture and then iteratively searches all the texture patches from the sampled texture space for a patch whose boundary best matches the current boundary of the synthesized texture. To control the number of the searching patches while preserving the seamless visual quality, we present two effective patch sampling techniques, namely uniformly distributed patch sampling and patch quantization clustering. The sampled texture patches will then be further optimized using principle component analysis. In this scheme, the complexity of the patch search in each step can be significantly reduced and thus the seamless texture synthesis can be performed in real-time. Experimental results show that the proposed schemes are of excellent efficiency for the visually seamless texture synthesis of a wide variety of textures ranging from regular to stochastic can be achieved.

Block coding is well known in the digital image coding literature. Vector quantization and transform coding are examples of well-known block coding techniques. Different images have many similar spatial blocks introducing inter-image similarity. The smaller the block size, the higher the inter-image similarity. In this paper, a new block coding algorithm based on inter-image similarity is proposed where it is claimed that any original image can be reconstructed from the blocks of any other image. The proposed algorithm is simply a vector quantization without the need to a codebook design algorithm and using matrix operations-based fast full search algorithm to find the local minimum root-mean-square error distortion measure to find the most similar code block to the input block. The proposed algorithm is applied in both spatial and transform domains with adaptive code block size. In the spatial domain, the encoding process has fidelity as high as 36.07dB with bit rate of 2.22bpp, while in the transform domain, the encoded image has good fidelity of 34.94dB with bit rate as low as 0.72bpp on the average. Moreover, the code image can be used as a secret key to provide secure communications.

In the present digital era, the exploitation of medical technologies and massive generation of medical data using different imaging modalities, adequate storage, management, and transmission of biomedical images necessitate image compression techniques. Vector quantization (VQ) is an effective image compression approach, and the widely employed VQ technique is Linde–Buzo–Gray (LBG), which generates local optimum codebooks for image compression. The codebook construction is treated as an optimization issue solved with utilization of metaheuristic optimization techniques. In this view, this paper designs an effective biomedical image compression technique in the cloud computing (CC) environment using Harris Hawks Optimization (HHO)-based LBG techniques. The HHO-LBG algorithm achieves a smooth transition among exploration as well as exploitation. To investigate the better performance of the HHO-LBG technique, an extensive set of simulations was carried out on benchmark biomedical images. The proposed HHO-LBG technique has accomplished promising results in terms of compression performance and reconstructed image quality.