Vector Quantization Method Research Articles

Prediction Transform GMM Vector Quantization for Wideband LSFs

Classified vector quantization (VQ) using a Gaussian Mixture Model (GMM) preclassifier is a powerful VQ method. A prediction-based lower-triangular transform (PLT) is proposed for the enhancement of VQ in each cluster. The PLT is defined for generic vector spaces in the context of the covariance method of linear prediction (LP). Optimal quantizer banks are designed in minimum noise structures whose codebooks are used for the proposed Cartesian split VQ (CSVQ), which improves their coding gain. CSVQ is tested forline spectral frequency (LSF) quantization of wideband speech spectra, revealing a comparable average performance to the Karhunen-Loeve transform (KLT) at lower rates with reduced outlier generation and computational complexity.

Application of Electronic Nose for Classification of Fish Sauces Using Transient Responses and Learning Vector Quantization Algorithm

Fish sauce is one of the signature condiments in various cuisines in many countries. In this paper, fish sauces are successfully classified into groups depending on their quality indicated by the level of total nitrogen content. We introduce an electronic nose technology together with a neural network algorithm to the classification of fish sauces. The transient responses are used as features for the creation of pattern vectors for odor samples. The result of principal component analysis shows well-separated patterns of fish sauce. Furthermore, we also apply the learning vector quantization method for the classification. As a result, we obtain high accuracy of more than 90% in the classification of fish sauce based on the level of total nitrogen content.

UL-Isomap based nonlinear dimensionality reduction for hyperspectral imagery classification

The paper proposes an upgraded landmark-Isometric mapping (UL-Isomap) method to solve the two problems of landmark selection and computational complexity in dimensionality reduction using landmark Isometric mapping (LIsomap) for hyperspectral imagery (HSI) classification. First, the vector quantization method is introduced to select proper landmarks for HSI data. The approach considers the variations in local density of pixels in the spectral space. It locates the unique landmarks representing the geometric structures of HSI data. Then, random projections are used to reduce the bands of HSI data. After that, the new method incorporates the Recursive Lanczos Bisection (RLB) algorithm to construct the fast approximate k-nearest neighbor graph. The RLB algorithm accompanied with random projections improves the speed of neighbor searching in UL-Isomap. After constructing the geodesic distance graph between landmarks and all pixels, the method uses a fast randomized low-rank approximate method to speed up the eigenvalue decomposition of the inner-product matrix in multidimensional scaling. Manifold coordinates of landmarks are then computed. Manifold coordinates of non-landmarks are computed through the pseudo inverse transformation of landmark coordinates. Five experiments on two different HSI datasets are run to test the new UL-Isomap method. Experimental results show that UL-Isomap surpasses LIsomap, both in the overall classification accuracy (OCA) and in computational speed, with a speed over 5 times faster. Moreover, the UL-Isomap method, when compared against the Isometric mapping (Isomap) method, obtains only slightly lower OCAs.

Multi-features-based license Plate Detection in Nighttime Environment

We propose a novel method for license plate detection in nighttime videos with complicated background scenes. The proposed method use multi-features-based combination way in the residual edge image. Our work firstly is color space conversion from RGB to La*b* color space and get enhancement L component of license plate image, then extract out multi-features of license plate, thirdly, multi-features combination method is used to get license plate detection, finally, search the license plate localization and segment the characters from the original nighttime environment images by using nonlinear vector quantization method. The experimental results show that the proposed method has more robust to interference characters and more accurate when compared with other methods.

Development an Automatic Translate Voice to Sign Language Animation Based-on MFCC and Vector Quantization Method

In this paper, we propose design and initial implementation of a robust system which can automatically translates voice into text and text to sign language animations. Sign Language Translation Systems could significantly improve deaf lives especially in communications, exchange of information and employment of machine for translation conversations from one language to another has. Therefore, considering these points, it seems necessary to study the speech recognition. Usually, the voice recognition algorithms address three major challenges. The first is extracting feature form speech and the second is when limited sound gallery are available for recognition, and the final challenge is to improve speaker dependent to speaker independent voice recognition. Extracting feature form speech is an important stage in our method. Different procedures are available for extracting feature form speech. One of the commonest of which used in speech recognition systems is Mel-Frequency Cepstral Coefficients (MFCCs). The algorithm starts with preprocessing and signal conditioning. Next extracting feature form speech using Cepstral coefficients will be done. Then the result of this process sends to segmentation part. Finally recognition part recognizes the words and then converting word recognized to facial animation. The project is still in progress and some new interesting methods are described in the current report.

A novel hierarchical decomposition vector quantization method for high-order LPC parameters

The paper investigates vector quantization coding of high-order (e.g., 20th-50th order) linear prediction coding (LPC) parameters, and proposes a novel hierarchical decomposition vector quantization method for a scalable speech coding framework with variable orders of LPC analysis. Instead of vector quantizing the whole group of LPC parameters in the linear spectral frequency (LSF) domain directly, the proposed method decomposes the high-order LPC model into several low-order (e.g., 10th-order) LPC models, and vector quantizes them in the LSF domain separately. For the decomposition, the high-order LPC model is converted into a group of reflection coefficients at first, and then the group is split into several subgroups and converted into multiple low-order LPC models. It is shown that the proposed method is naturally suitable for a scalable coding framework where the information of the decomposed low-order LPC models can be encoded into a multi-layered bitstream and can be combined in a progressive way to recover the high-order LPC information. Experiments in a scalable coding framework with variable LPC analysis orders (10-50) reveal that, compared to a direct vector quantization scheme, the proposed method can reduce the size of the codebook and the number of coding bits significantly, and can also efficiently reduce the computation cost.

Design of Brain-Machine Interface Using Near-Infrared Spectroscopy

In recent years, the fields of robotics and medical science have been paying close attention to brainmachine interface (BMI) systems. BMI observes human cerebral activity and use the collected data as the input to various instruments. If such a systemcould be effectively realized, it could be used as a new intuitive input interface for application to human-robot interactions, welfare scenarios, etc. In this paper, we discussed a design problem related to a BMI system using near-infrared spectroscopy (NIRS). We developed a brain state classifier based on the learning vector quantization (LVQ) method. The proposed method classifies the cerebral blood flow patterns and outputs the brain state estimate. The classification experiments showed that the proposed method can successfully classify not only human physical motions and motor imageries, but also human emotions and human mental commands issued to a robot. Especially, in the classification of “the mental commands to a robot,” we successfully realized the imagery classification of five different mental commands. The results point to the potential of NIRS-based brain machine interfaces.

An image segmentation method based on the fusion of vector quantization and edge detection with applications to medical image processing

In the existing segmentation algorithms, most of them take single pixel as processing unit and segment an image mainly based on the gray value information of the image pixels. However, the spatially structural information between pixels of an image provides even more important information of the image. In order to effectively exploit both the gray value and the spatial information of image pixels, this paper proposes a fusion method for image segmentation by jointly utilizing vector quantization and edge detection methods. In the method, the image to be segmented is divided into small sub-blocks with each sub-block constituting a vector and the vectors are classified into two patterns, called the edge pattern and non-edge pattern, by using an edge detection algorithm. The image is then processed further, in which a Boundary Detection (BD) algorithm is developed for extracting the refined boundary curves in the edge pattern vectors and a Vector Quantization (VQ) approach is presented for segmenting the non-edge pattern vectors. In addition, an SOM neural network is proposed for realizing the VQ algorithm adaptively. Finally, a fusion scheme is designed to synthesize the results of VQ and BD to accomplish the segmentation. Simulation experiments and comparison studies have been conducted with applications to medical image processing in the paper, and the results validate the effectiveness of the proposed method.

APPLICATION OF NEURAL NETWORKS ON BLOOD SERUM IMAGE FOR EARLY DETECTION OF TYPHUS

Background: Typhus is a disease caused by Salmonella typhi, Salmonella paratyphi A Salmonella parathypi B, dan Salmonella paratyphi C bacteria that attacks digestive tract and caused infection in small intestine. The common test that performed in the laboratory is widal test. The result reading of the widal test still processed manually with looking the turbidity caused by the agglutination. Aim: The research was made to decrease human error by creating a program based on artificial neural network (ANN) with learning vector quantization (LVQ) method. Method: Input of this program is image of blood serum that has reacted with widal reagen. Image procesing start with grayscaling, filtering, and thresholding. Result: Output of this program is divided into two classes, normal and typhus detected. Conclusion: From this experiment result that using 24 testing data, gives the accuracy of this program 95.833% with 1 error result from 24 testing data.

Unsupervised language model adaptation for handwritten Chinese text recognition

This paper presents an effective approach for unsupervised language model adaptation (LMA) using multiple models in offline recognition of unconstrained handwritten Chinese texts. The domain of the document to recognize is variable and usually unknown a priori, so we use a two-pass recognition strategy with a pre-defined multi-domain language model set. We propose three methods to dynamically generate an adaptive language model to match the text output by first-pass recognition: model selection, model combination and model reconstruction. In model selection, we use the language model with minimum perplexity on the first-pass recognized text. By model combination, we learn the combination weights via minimizing the sum of squared error with both L2-norm and L1-norm regularization. For model reconstruction, we use a group of orthogonal bases to reconstruct a language model with the coefficients learned to match the document to recognize. Moreover, we reduce the storage size of multiple language models using two compression methods of split vector quantization (SVQ) and principal component analysis (PCA). Comprehensive experiments on two public Chinese handwriting databases CASIA-HWDB and HIT-MW show that the proposed unsupervised LMA approach improves the recognition performance impressively, particularly for ancient domain documents with the recognition accuracy improved by 7 percent. Meanwhile, the combination of the two compression methods largely reduces the storage size of language models with little loss of recognition accuracy.

MEDICAL IMAGE COMPRESSION TECHNIQUE USING LISTLESS SET PARTITIONING IN HIERARCHICAL TREES AND CONTEXTUAL VECTOR QUANTIZATION FOR BRAIN IMAGES

A hybrid image compression techniques has been developed to compress medical images. Due to the extensive use of medical images like CT and MR scan, these medical imagery are stored for a longer period for th e continuous monitoring of the patients and the amount of data associated with images is large and it oc cupies enormous storage capacity. So, the medical images need to be compressed to reduce the storage cost and for transmission without any loss. In this study, a hyb rid method which combines the Listless Set Partitio ning in Hierarchical Trees (LSPIHT) and the Contextual Vector Quantization (CVQ) method for the compression of brain images. Here, the region containing the most important information for diagnosis is called Regio n of Interest (ROI) and this is to be compressed with ou t any loss in the quality. In this method, the ROI is encoded separately using LSPIHT and the Back Ground region (BG) is encoded using CVQ. Finally, the two regions are merged together to get the reconstructed image. Our results show that the proposed method gives ve ry good image quality for diagnosis without any degrad able loss. The performance of the compression techn ique is evaluated using the parameters (CR, MSE, PSNR) and achieved better result compared to other existin g methods. As a result, we strongly believe that usin g our method, we can overcome the limitations in st orage and transmission of medical images that are produce d day by day.

Quantized kernel recursive least squares algorithm.

In a recent paper, we developed a novel quantized kernel least mean square algorithm, in which the input space is quantized (partitioned into smaller regions) and the network size is upper bounded by the quantization codebook size (number of the regions). In this paper, we propose the quantized kernel least squares regression, and derive the optimal solution. By incorporating a simple online vector quantization method, we derive a recursive algorithm to update the solution, namely the quantized kernel recursive least squares algorithm. The good performance of the new algorithm is demonstrated by Monte Carlo simulations.

Gases/Odors Identification With Artificial Immune Recognition System Using Thick Film Gas Sensor Array Responses

This paper discusses the robustness of the artificial immune recognition system (AIRS) for the gases/odors identification problem. The steady state responses of a thick-film sensor array with four sensor elements with exposure of four gases, viz., H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> , CO, CH <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</sub> , and LPG, are used as input data. The AIRS algorithm with its versions including AIRS1, AIRS2, and parallel AIRS is applied to classify the unseen gases/odors data with duly trained networks. The classification accuracy of the AIRS algorithm is compared with radial basis function neural network, naive bayes, and learning vector quantization methods. The results obtained with the AIRS are found more promising in this experiment. The results are verified using a cross-validation scheme.

Robust Track-Following Controller Design for Hard Disk Drives With Irregular Sampling

This paper considers robust controller design for track-following in hard disk drives (HDD) with irregular sampling of the position error signal (PES) but regular (clock-driven) control updates. This sampling and actuation behavior is modeled by applying a novel discretization method to a continuous-time model of an HDD, resulting in a discrete-time linear periodically time-varying model. Then, the controller design is performed using optimal <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">H</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">∞</sub> control for periodic systems and uses a generalization of the disk margin to quantify the robustness of the closed-loop system. To show the effectiveness of the proposed method, the design methodology is applied to a hard disk drive model and the resulting controller is validated by examining its nominal performance in terms of the root mean square of the standard deviation of the PES and robustness in terms of disk margin. Since the proposed controller has too many parameters to be implementable on an HDD due to memory limitations, we use a vector quantization method to approximate the entire parameter set of the designed controller by a smaller set of parameters.

Optimized Product Quantization.

Product quantization (PQ) is an effective vector quantization method. A product quantizer can generate an exponentially large codebook at very low memory/time cost. The essence of PQ is to decompose the high-dimensional vector space into the Cartesian product of subspaces and then quantize these subspaces separately. The optimal space decomposition is important for the PQ performance, but still remains an unaddressed issue. In this paper, we optimize PQ by minimizing quantization distortions w.r.t the space decomposition and the quantization codebooks. We present two novel solutions to this challenging optimization problem. The first solution iteratively solves two simpler sub-problems. The second solution is based on a Gaussian assumption and provides theoretical analysis of the optimality. We evaluate our optimized product quantizers in three applications: (i) compact encoding for exhaustive ranking [1], (ii) building inverted multi-indexing for non-exhaustive search [2], and (iii) compacting image representations for image retrieval [3]. In all applications our optimized product quantizers outperform existing solutions.

Fast compression of computer-generated holographic images based on a GPU-accelerated skip-dimension vector quantization method

A method for fast and low bit-rate compression of digital holograms based on a new vector quantization (VQ) method known as the skip-dimension VQ (SDVQ) is proposed. Briefly, a complex hologram is converted into a real off-axis hologram, and partitioned into a set of image vectors. The image vectors are passed into a graphic processing unit (GPU), and compressed through SDVQ into a set of code indices considerably smaller in data size than the source hologram. Experimental evaluation reveals that our scheme is capable of compressing a digital hologram to a compression ratio of over 500 times, in approximately 20-22 ms.

Development of a brain MRI-based hidden Markov model for dementia recognition

BackgroundDementia is an age-related cognitive decline which is indicated by an early degeneration of cortical and sub-cortical structures. Characterizing those morphological changes can help to understand the disease development and contribute to disease early prediction and prevention. But modeling that can best capture brain structural variability and can be valid in both disease classification and interpretation is extremely challenging. The current study aimed to establish a computational approach for modeling the magnetic resonance imaging (MRI)-based structural complexity of the brain using the framework of hidden Markov models (HMMs) for dementia recognition.MethodsRegularity dimension and semi-variogram were used to extract structural features of the brains, and vector quantization method was applied to convert extracted feature vectors to prototype vectors. The output VQ indices were then utilized to estimate parameters for HMMs. To validate its accuracy and robustness, experiments were carried out on individuals who were characterized as non-demented and mild Alzheimer's diseased. Four HMMs were constructed based on the cohort of non-demented young, middle-aged, elder and demented elder subjects separately. Classification was carried out using a data set including both non-demented and demented individuals with a wide age range.ResultsThe proposed HMMs have succeeded in recognition of individual who has mild Alzheimer's disease and achieved a better classification accuracy compared to other related works using different classifiers. Results have shown the ability of the proposed modeling for recognition of early dementia.ConclusionThe findings from this research will allow individual classification to support the early diagnosis and prediction of dementia. By using the brain MRI-based HMMs developed in our proposed research, it will be more efficient, robust and can be easily used by clinicians as a computer-aid tool for validating imaging bio-markers for early prediction of dementia.

An Intelligent Method Based Medical Image Compression

Compression methods are important in many medical applications to ensure fast interactivity through large sets of images (e.g. volumetric data sets, image databases), for searching context dependant images and for quantitative analysis of measured data. Medical data are increasingly represented in digital form. The limitations in transmission bandwidth and storage space on one side and the growing size of image datasets on the other side has necessitated the need for efficient methods and tools for implementation. Many techniques for achieving data compression have been introduced. Wavelet transform techniques currently provide the most promising approach to high-quality image compression, which is essential for Teleradiology. This paper presents an approach of intelligent method to design a vector quantizer for image compression. The image is compressed without any loss of information. It also provides a comparative study in the view of simplicity, storage space, robustness and transfer time of various vector quantization methods. The proposed approach presents an efficient method of vector quantization for image compression and application of SOFM.

Image Reconstruction Using Discrete Wavelet Transform

In the recent growth of data intensive and multimedia based applications, efficient image compression solutions are becoming critical. The main objective of Image Compression is to reduce redundancy of the data and improve the efficiency. The main techniques used are Fourier Analysis, Discrete Cosine Transform vector quantization method, sub-band coding method. The drawbacks in the above methods are, they cannot be used for real time systems. In order to overcome these problems, the Wavelet Transform method has been introduced. Wavelet Analysis is highly capable of revealing aspects of data like trends, breakdown points, discontinuities in higher derivates and self similarity and can often compress or diagnose a signal without appreciable degradation. Here, we implement a lossy image compression technique using Matlab Wavelet Toolbox and Matlab Functions where the wavelet transform of the signal is performed, then calculated a threshold based on the compression ratio acquired by the user.

Random and system errors in nutrient analysis: An application of adaptive neural-network protocols

USDA’s nutritional reference database is the primary database used by several nutrient analyses software. Despite the use of one database, variations among systems emerge impacting nutrient estimation. Consequently, this paper explores these critical differences by classifying sources of variations leading to computational errors. Learning Vector Quantization and Bayesian (neural-network) methods were applied to characterize pattern of errors. As a result, two common errors were identified to be prevalent in nutrient analyses. Random errors (37.1% of total variance) stemming from selection process (Operator) because of the enormity of food items listed in the database and system errors (59.5% of total variance) due to measurements (units, tools and procedures) variations. Computerized classification of random and system errors are of significant interest in the field of nutrition and food engineering because, it provides practical solutions to database query and optimization of inherent variations due to measurement errors that are beneficial to research methodologies.