Abstract: 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.

Abstract: We introduce a new compression scheme for high-dimensional vectors that approximates the vectors using sums of M codewords coming from M different codebooks. We show that the proposed scheme permits efficient distance and scalar product computations between compressed and uncompressed vectors. We further suggest vector encoding and codebook learning algorithms that can minimize the coding error within the proposed scheme. In the experiments, we demonstrate that the proposed compression can be used instead of or together with product quantization. Compared to product quantization and its optimized versions, the proposed compression approach leads to lower coding approximation errors, higher accuracy of approximate nearest neighbor search in the datasets of visual descriptors, and lower image classification error, whenever the classifiers are learned on or applied to compressed vectors.

Abstract: In this paper, we propose a novel joint data-hiding and compression scheme for digital images using side match vector quantization (SMVQ) and image inpainting. The two functions of data hiding and image compression can be integrated into one single module seamlessly. On the sender side, except for the blocks in the leftmost and topmost of the image, each of the other residual blocks in raster-scanning order can be embedded with secret data and compressed simultaneously by SMVQ or image inpainting adaptively according to the current embedding bit. Vector quantization is also utilized for some complex blocks to control the visual distortion and error diffusion caused by the progressive compression. After segmenting the image compressed codes into a series of sections by the indicator bits, the receiver can achieve the extraction of secret bits and image decompression successfully according to the index values in the segmented sections. Experimental results demonstrate the effectiveness of the proposed scheme.

Abstract: We describe an implementation of the Deformable Parts Model [1] that operates in a user-defined time-frame. Our implementation uses a variety of mechanism to trade-off speed against accuracy. Our implementation can detect all 20 PASCAL 2007 objects simultaneously at 30Hz with an mAP of 0.26. At 15Hz, its mAP is 0.30; and at 100Hz, its mAP is 0.16. By comparison the reference implementation of [1] runs at 0.07Hz and mAP of 0.33 and a fast GPU implementation runs at 1Hz. Our technique is over an order of magnitude faster than the previous fastest DPM implementation. Our implementation exploits a series of important speedup mechanisms. We use the cascade framework of [3] and the vector quantization technique of [2]. To speed up feature computation, we compute HOG features at few scales, and apply many interpolated templates. A hierarchical vector quantization method is used to compress HOG features for fast template evaluation. An object proposal step uses hash-table methods to identify locations where evaluating templates would be most useful; these locations are inserted into a priority queue, and processed in a detection phase. Both proposal and detection phases have an any-time property. Our method applies to legacy templates, and no retraining is required.

Abstract: 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.

Abstract: Color quantization is the process of reducing the number of colors in a digital image. The main objective of quantization process is that significant information should be preserved while reducing the color of an image. In other words, quantization process shouldn't cause significant information loss in the image. In this paper, a short review of color quantization is presented and a new color quantization method based on artificial bee colony algorithm (ABC) is proposed. The performance of the proposed method is evaluated by comparing it with the performance of the most widely used quantization methods such as K-means, Fuzzy C Means (FCM), minimum variance and particle swarm optimization (PSO). The obtained results indicate that the proposed method is superior to the others.

Abstract: Predictive coding is attractive for compression onboard of spacecrafts thanks to its low computational complexity, modest memory requirements and the ability to accurately control quality on a pixel-by-pixel basis. Traditionally, predictive compression focused on the lossless and near-lossless modes of operation where the maximum error can be bounded but the rate of the compressed image is variable. Rate control is considered a challenging problem for predictive encoders due to the dependencies between quantization and prediction in the feedback loop, and the lack of a signal representation that packs the signal's energy into few coefficients. In this paper, we show that it is possible to design a rate control scheme intended for onboard implementation. In particular, we propose a general framework to select quantizers in each spatial and spectral region of an image so as to achieve the desired target rate while minimizing distortion. The rate control algorithm allows to achieve lossy, near-lossless compression, and any in-between type of compression, e.g., lossy compression with a near-lossless constraint. While this framework is independent of the specific predictor used, in order to show its performance, in this paper we tailor it to the predictor adopted by the CCSDS-123 lossless compression standard, obtaining an extension that allows to perform lossless, near-lossless and lossy compression in a single package. We show that the rate controller has excellent performance in terms of accuracy in the output rate, rate-distortion characteristics and is extremely competitive with respect to state-of-the-art transform coding.

Abstract: Learning vector quantization applying non-standard metrics became quite popular for classification performance improvement compared to standard approaches using the Euclidean distance. Kernel metrics and quadratic forms belong to the most promising approaches. In this paper we consider Minkowski distances (lp-norms). In particular, l1-norms are known to be robust against noise in data, such that, if this structural knowledge is available in advance about the data, this norm should be utilized. However, application in gradient based learning algorithms based on distance evaluations need to calculate the respective derivatives. Because lp-distance formulas contain the absolute approximations thereof are required. We consider in this paper several approaches for smooth consistent approximations for numerical evaluations and demonstrate the applicability for exemplary real world applications.

Abstract: In this research, we propose Projected Residual Vector Quantization (PRVQ) to deal with the problem of large-scale approximate nearest neighbor (ANN) search in a high-dimensional space. A lot of quantization-based ANN search algorithms have been proposed in the past few years. However, most of the existing methods discard the projection errors generated in the dimension reduction process, which inevitably decreases the search accuracy. In view of that, the authors propose a method of projected residual vector quantization for ANN search that considers the projection errors in the quantization process. They also design three simple and effective optimization strategies to improve the performance of the PRVQ algorithm. The authors have integrated the proposed PRVQ algorithm into a mobile landmark recognition system to prove its effectiveness.

Abstract: Acoustic event detection is an important step for audio content analysis and retrieval. Traditional detection techniques model the acoustic events on frame-based spectral features. Considering the temporal-frequency structures of acoustic events may be distributed in time-scales beyond frames, we propose to represent those structures as a bag of spectral patch exemplars. In order to learn the representative exemplars, k-means clustering based vector quantization (VQ) was applied on the whitened spectral patches which makes the learned exemplars focus on high-order statistical structure. With the learned spectral exemplars, a sparse feature representation is extracted based on the similarity measurement to the learned exemplars. A support vector machine (SVM) classifier was built on the sparse representation for acoustic event detection. Our experimental results showed that the sparse representation based on the patch based exemplars significantly improved the performance compared with traditional frame based representations.

Abstract: Following a nonparametric approach, we suggest a time-series clustering method. Our clustering approach combines the benefits connected to the interpretative power of the nonparametric representation of the time series, and the clustering and vector quantization informational gain produced by the adopted unsupervised neural networks technique, enhanced with the self-organizing maps ordering and topological preservation abilities. The proposed clustering method takes into account a composite wavelet-based information of the multivariate time series by adding to the information connected to the wavelet variance, namely the influence of variability of individual univariate components of the multivariate time series across scales, the information associated to wavelet correlation, represented by the interaction between pairs of univariate components of the multivariate time series at each scale, and then suitably tuning the combination of these pieces of information. In order to assess the effectiveness of the proposed clustering approach, a simulation study and an empirical application are shown. Copyright © 2013 John Wiley & Sons, Ltd.

Abstract: In this paper, a new hierarchical color quantization method based on self-organizing maps that provides different levels of quantization is presented. Color quantization (CQ) is a typical image processing task, which consists of selecting a small number of code vectors from a set of available colors to represent a high color resolution image with minimum perceptual distortion. Several techniques have been proposed for CQ based on splitting algorithms or cluster analysis. Artificial neural networks and, more concretely, self-organizing models have been usually utilized for this purpose. The self-organizing map (SOM) is one of the most useful algorithms for color image quantization. However, it has some difficulties related to its fixed network architecture and the lack of representation of hierarchical relationships among data. The growing hierarchical SOM (GHSOM) tries to face these problems derived from the SOM model. The architecture of the GHSOM is established during the unsupervised learning process according to the input data. Furthermore, the proposed color quantizer allows the evaluation of different color quantization rates under different codebook sizes, according to the number of levels of the generated neural hierarchy. The experimental results show the good performance of this approach compared to other quantizers based on self-organization.

Abstract: Human recognition is an essential requirement for human-centric surveillance, activity recognition, gait recognition etc. Inaccurate recognition of humans in such applications may leads to false alarm and unnecessary computation. In the proposed work a robust background modeling algorithm using fuzzy logic is used to detect foreground objects. Three distinct features are extracted from the contours of detected objects. An unique aggregated feature vector is formed using a fuzzy inference system by aggregating three feature vectors. To minimize computation in recognition using Hidden Markov model (HMM), the length of final feature vector is reduced using vector quantization. The proposed method is explained using five basic phases; background modeling and foreground object detection, features extraction, aggregated feature vector calculation, vector quantization, and recognition using Hidden Markov model.

Abstract: The top-performing systems for billion-scale high-dimensional approximate nearest neighbor (ANN) search are all based on two-layer architectures that include an indexing structure and a compressed datapoints layer. An indexing structure is crucial as it allows to avoid exhaustive search, while the lossy data compression is needed to fit the dataset into RAM. Several of the most successful systems use product quantization (PQ) for both the indexing and the dataset compression layers. These systems are however limited in the way they exploit the interaction of product quantization processes that happen at different stages of these systems. Here we introduce and evaluate two approximate nearest neighbor search systems that both exploit the synergy of product quantization processes in a more efficient way. The first system, called Fast Bilayer Product Quantization (FBPQ), speeds up the runtime of the baseline system (Multi-D-ADC) by several times, while achieving the same accuracy. The second system, Hierarchical Bilayer Product Quantization (HBPQ) provides a significantly better recall for the same runtime at a cost of small memory footprint increase. For the BIGANN dataset of billion SIFT descriptors, the 10% increase in Recall@1 and the 17% increase in Recall@10 is observed.

Abstract: We study joint source-channel coding (JSCC) of compressed sensing (CS) measurements using vector quantizer (VQ). We develop a framework for realizing optimum JSCC schemes that enable encoding and transmitting CS measurements of a sparse source over discrete memoryless channels, and decoding the sparse source signal. For this purpose, the optimal design of encoder-decoder pair of a VQ is considered, where the optimality is addressed by minimizing end-to-end mean square error (MSE). We derive a theoretical lower-bound on the MSE performance, and propose a practical encoder-decoder design through an iterative algorithm. The resulting coding scheme is referred to as channel- optimized VQ for CS, coined COVQ-CS. In order to address the encoding complexity issue of the COVQ-CS, we propose to use a structured quantizer, namely low complexity multi-stage VQ (MSVQ). We derive new encoding and decoding conditions for the MSVQ, and then propose a practical encoder-decoder design algorithm referred to as channel-optimized MSVQ for CS, coined COMSVQ-CS. Through simulation studies, we compare the proposed schemes vis-a-vis relevant quantizers.

Abstract: In recent years, several researches are being done to improve the means by which human to machine interaction. With the development of input devices like keyboard, mouse and pen are not sufficient due to this limitation direct use of hand gesture as an input device to provide natural human to machine interaction. The objective of this paper is to implement the vision based hand gesture recognition system to control the movement of robot. We can use of Scale invariant feature transform (SIFT) for extract the keypoint from the gesture image capture by single sensing device. Space incompatibility of SIFT keypoint causes bag of feature approach was introduced. Then use the vector quantization will map the keypoint extracted from SIFT into unified dimensional histogram vector after the K-mean clustering. The histogram vectors as an input to multiclass SVM classifier for recognize the gesture. Generate the grammar apply to the robot to control the movements (Left, Right, Straight ward, Backward, stop) of robot.

Abstract: We present an introductory survey to optimal vector quantization and its first applications to Numerical Probability and, to a lesser extent to Information Theory and Data Mining. Both theoretical results on the quantization rate of a random vector taking values in R^d (equipped with the canonical Euclidean norm) and the learning procedures that allow to design optimal quantizers (CLVQ and Lloyd's I procedures) are presented. We also introduce and investigate the more recent notion of {\em greedy quantization} which may be seen as a sequential optimal quantization. A rate optimal result is established. A brief comparison with Quasi-Monte Carlo method is also carried out.

Abstract: recognition is the process of recognizing the speaker based on characteristics such as pitch ,tone in the speech wave.Background noise influences the overall efficiency of speaker recognition system and is still considered as one of the most challenging issue in Speaker Recognition System (SRS). In this paper mel-frequency cepstral coefficients (MFCC) feature is used along with Vector Quantisation(VQ)- LBG (Linde, Buzo and Gray, 1980) algorithm for designing SRS. MFCC feature is extracted from the input speech and then vector quantization of the extracted MFCC features is done using VQLBG algorithm. It reduces the dimensionality of the input vector .These MFCCs are used as the speaker features for matching via Support Vector Machine (SVM) method. The experimental results show that the proposed text- dependent speaker identification system gives an accuracy rate of 95.0%.

Abstract: Enhancing speech recognition is the primary intention of this work. In this paper a novel speech recognition method based on vector quantization and improved particle swarm optimization (IPSO) is suggested. The suggested methodology contains four stages, namely, (i) denoising, (ii) feature mining (iii), vector quantization, and (iv) IPSO based hidden Markov model (HMM) technique (IP-HMM). At first, the speech signals are denoised using median filter. Next, characteristics such as peak, pitch spectrum, Mel frequency Cepstral coefficients (MFCC), mean, standard deviation, and minimum and maximum of the signal are extorted from the denoised signal. Following that, to accomplish the training process, the extracted characteristics are given to genetic algorithm based codebook generation in vector quantization. The initial populations are created by selecting random code vectors from the training set for the codebooks for the genetic algorithm process and IP-HMM helps in doing the recognition. At this point the creativeness will be done in terms of one of the genetic operation crossovers. The proposed speech recognition technique offers 97.14% accuracy.

Abstract: In recent years, facial expression recognition has become a useful scheme for computers to affectively understand the emotional state of human beings. Facial representation and facial expression recognition under unconstrained environments have been two critical issues for facial expression recognition systems. This thesis contributes to the research and development of facial expression recognition systems from two aspects: first, feature extraction for facial expression recognition, and second, applications to challenging conditions. Spatial and temporal feature extraction methods are introduced to provide effective and discriminative features for facial expression recognition. The thesis begins with a spatial feature extraction method. This descriptor exploits magnitude while it improves local quantized pattern using improved vector quantization. It also makes the statistical patterns domain-adaptive and compact. Then, the thesis discusses two spatiotemporal feature extraction methods. The first method uses monogenic signal analysis as a preprocessing stage and extracts spatiotemporal features using local binary pattern. The second method extracts sparse spatiotemporal features using sparse cuboids and spatiotemporal local binary pattern. Both methods increase the discriminative capability of local binary pattern in the temporal domain. Based on feature extraction methods, three practical conditions, including illumination variations, facial occlusion and pose changes, are studied for the applications of facial expression recognition. First, with near-infrared imaging technique, a discriminative component-based single feature descriptor is proposed to achieve a high degree of robustness and stability to illumination variations. Second, occlusion detection is proposed to dynamically detect the occluded face regions. A novel system is further designed for handling effectively facial occlusion. Lastly, multiview discriminative neighbor preserving embedding is developed to deal with pose change, which formulates multi-view facial expression recognition as a generalized eigenvalue problem. Experimental results on publicly available databases show that the effectiveness of the proposed approaches for the applications of facial expression recognition.

Abstract: We study joint source-channel coding (JSCC) of compressed sensing (CS) measurements using vector quantizer (VQ). We develop a framework for realizing optimum JSCC schemes that enable encoding and transmitting CS measurements of a sparse source over discrete memoryless channels, and decoding the sparse source signal. For this purpose, the optimal design of encoder-decoder pair of a VQ is considered, where the optimality is addressed by minimizing end-to-end mean square error (MSE). We derive a theoretical lower bound on the MSE performance and propose a practical encoder-decoder design through an iterative algorithm. The resulting coding scheme is referred to as channel-optimized VQ for CS, coined COVQ-CS. In order to address the encoding complexity issue of the COVQ-CS, we propose to use a structured quantizer, namely low-complexity multistage VQ (MSVQ). We derive new encoding and decoding conditions for the MSVQ and then propose a practical encoder–decoder design algorithm referred to as channel-optimized MSVQ for CS, coined COMSVQ-CS. Through simulation studies, we compare the proposed schemes vis-a-vis relevant quantizers.