Search

In this paper, a novel method of locating the sun spot is presented. Used the math transform domain algorithm to emphasize the brightness of the sun area in the fisheye picture, the human's vision brightness sensitivity is simulated by this optic color filter impact. The small sun in the fisheye picture is segmented, and transformed to the plane picture instead of the whole picture. It is easy to get the coordinates of the sun area barycenter from the plane picture to the fisheye picture. Then the azimuth angle and the vertical angle between the vision point and the sun spot are calculated. The results of the experimentation show that the amount of computation of the algorithm is reduced a lot. It is accurate, fast and real-time. [ABSTRACT FROM AUTHOR]

Simulation optimization techniques are discussed for multi-chain Markov decision processes (MDPs) by the learning of performance potentials. Different from ergodic or unichain models, where a single sample path suffices to be used for the learning of potentials, under a multichain case, there are more than one recurrent classes for the underlying Markov chain, therefore the sample path has to be restarted often so as not to circulate only in one recurrent class. Similar to unichain models, temporal difference (TD) learning algorithms can also be developed for learning potentials. In addition, by representing the estimates of potentials via a neural network, one neuro-dynamic programming (NDP) method, i.e., the critic algorithm, is derived as what has been supposed for unichain models. The obtained results are also applicable for general multichain semi-Markov decision processes (SMDPs), and we use a numerical example to illustrate the extension. [ABSTRACT FROM AUTHOR]

Wiener-typed nonlinear systems with hard input constraints are ubiquitous in industrial processes. However, because of the complex structure, there are only few achievements on the control algorithm for constrained Wiener-typed system. An improved dual-mode control algorithm is put forward. Firstly, Zeroin Algorithm is applied to obtain the inverse of the static nonlinear block of the Wiener-typed nonlinear system. Then, we define the invariant ellipsoid sets for estimated state and estimated error respectively, and guarantee the feasibility, stability and convergence of this algorithm by the theory of invariant set combined with Dual-Model approach[1]. In contrast to traditional algorithms, this one has the advantages of larger initial stable region in state space and higher tracking accuracy. Finally, the feasibility and superiority of the proposed algorithm are validated by case studies. [ABSTRACT FROM AUTHOR]

This paper proposes an effective Particle Swarm Optimization (PSO) based Memetic Algorithm (MA) for Traveling Salesman Problem (TSP) which is a typical NP-hard combinatorial optimization problem with strong engineering background. In the proposed PSO-based MA (PSOMA), a novel encoding scheme is developed, and an effective local search based on Simulated Annealing (SA) with adaptive meta-Lamarckian learning strategy is proposed and incorporated into PSO. Simulation results based on well-known benchmarks and comparisons with some existing algorithms demonstrate the effectiveness of the proposed hybrid algorithm for TSP. [ABSTRACT FROM AUTHOR]

A mobile manipulator-a serial connection of a mobile robot and task robot- has the abilities of both moving and performing a task. In this paper, we show that a mobile manipulator transfers an arbitrary object. Here, we figure out the solution for the optimal configuration of a mobile manipulator with a series of tasks. For this, first we define the task vector for contiguous tasks. And after making the major(longest principal) axis of manipulability ellipsoid generated by a task robot to correspond to the task vector, we can find the optimal posture of a mobile manipulator by considering the posture of a task robot for the configuration of a task conversion and computing the position of the mobile robot kinematics. In addition to that, we try to decrease the distance error and direction error of a mobile robot using the Lyapunov equation and focus our mind on the control of a task robot using the inverse kinematics equation. [ABSTRACT FROM AUTHOR]

This paper considers a single machine scheduling problem with sequence-dependent setup times to minimize the maximum lateness. A genetic algorithm is developed in which an effective binary coding based on the problem properties is presented, and a heuristic for sequencing the batches given the batching structure is proposed. Computational experiments show that the proposed genetic algorithm performs well in solving the problem, and is capable of effectively solving large problems involving 400 jobs. [ABSTRACT FROM AUTHOR]

This paper proposes an effective Differential Evolution (DE) based hybrid algorithm for Multi-objective Permutation Flow Shop Scheduling Problem (MPFSSP), which is a typical NP-hard combinatorial optimization problem. In the proposed Multi-objective Hybrid DE (MOHDE), both DE-based searching operators and some special local searching operators are designed to balance the exploration and exploitation abilities. Firstly, to make DE suitable for solving MPFSSP, a largest-order-value (LOV) rule based on random key representation is developed to convert the continuous values of individuals in DE to job permutations. Then, to enrich the searching behaviors and to avoid premature convergence, a Variable Neighborhood Search (VNS) based local search with multiple different neighborhoods is designed and incorporated into the MOHDE. Simulation results and comparisons with the famous random-weight genetic algorithm (RWGA) demonstrate the effectiveness and robustness of our proposed MOHDE. [ABSTRACT FROM AUTHOR]

A Mobile Manipulator-a serial connection of a mobile robot and a task robot-is redundant by itself. Using it's redundant freedom, a mobile manipulator can perform various task. In this paper, to improve task execution efficiency utilizing the redundancy, optimal configurations of the mobile manipulator are maintained while it is moving to a new task point. And using a cost function for optimality defined as a combination of the square errors of the desired and actual configurations of the mobile robot and of the task robot, the job which the mobile manipulator performs is optimized. Here, The proposed algorithm is experimentally verified and discussed with a mobile manipulator, PURL-II [ABSTRACT FROM AUTHOR]

An imitating-natural-reading paradigm was used to induce robust visual evoked potentials (VEPs) which as carriers for a brain-computer interface based mental speller. Support vector machine (SVM) was adopted in the single-trail classification on the features. To improve the accuracy of pattern recognition, thus to boost up the bit rate of whole system, a 300ms window was used to estimate the accurate time of target stimuli present from EEG signals. As the spontaneous EEG could be regarded as a stationary random process in a short period, a whiten filter was constructed by the AR parameters which calculated from those non-target induced signals. In succession, real-world signals were input to the filter where the spontaneous EEGs were whitened. Finally, a wavelet method was applied to have the white signals filtered. The results boosted up the classification accuracy by enhancing the target signals. [ABSTRACT FROM AUTHOR]

This paper investigates the estimation of fusion weights under varying acoustic noise conditions for audio-visual multi-level hybrid fusion strategy in speaker identification. The multi-level fusion combines model level and decision level fusion via dynamic Bayesian networks (DBNs). A novel methodology known as support vector regression (SVR) is utilized to estimate the fusion weights directly from audio features; Sigma-Pi network sampling method is also incorporated to reduce feature dimensions. Experiments on the homegrown Chinese database and CMU English database both demonstrate that the method improves the accuracies of audio-visual bimodal speaker identification under dynamically varying acoustic noise conditions. [ABSTRACT FROM AUTHOR]

This paper presents a system that accurately records an animal's motion during clinical experiments, using a camera connected to a computer, and analyzing and serving data. Using images input through a general CCD camera, the system separates background and animal, and stores the location and state. Existing systems support tracking of a single animal and analysis of path, velocity, and so on. However, the proposed system supports multiple animal tracking and analysis of animal conditions by modeling the shape of the animal. [ABSTRACT FROM AUTHOR]

Support vector machine (SVM) and v-SVM are novel type of learning machine, which have shown to provide better generalization performance than traditional techniques. This thesis introduces a new type of fuzzy support vector machine (Fv-SVM), which based on v-SVM. The new algorithm considers that the input samples have different contributions to the final result, so fuzzy memberships are used to determine the effects of input samples. It also discusses in detail the core algorithms determine the fuzzy memberships based on kernel methods. In the experiments Fv-SVM is used for character recognition. The results show that Fv-SVM has low error rate and better generalization ability. [ABSTRACT FROM AUTHOR]

In this paper, Statistical Neural Networks have been proven to be an effective classifier method for large sample and high dimensional letter recognition problem. For this purpose, Probabilistic Neural Network (PNN) and General Regression Neural Networks (GRNN) have been applied to classify the 26 capital letters in the English alphabet. Principal Component Analysis (PCA) has been established as a feature extraction and a data compression method to achieve less computational complexity. The low computational complexity obtained by PCA provides a solution for high dimensional letter recognition problem for online operations. Simulation results illustrate that GRNN and PNN are suitable and effective methods for solving classification problems with higher classification accuracy and better generalization performances than their counterparts. [ABSTRACT FROM AUTHOR]

The shape or outlines of planar barefoot impressions of persons has been found useful in many cases for personal identification. This paper performs an automatic analysis on the shape of barefoot impressions, totally 28 numerical feature measurements are extracted to describe the geometry and structure properties of the barefoot impression's shape. Each measurement on the footprint outlines extracted in this work is examined for the distinguishability and the consistence by its population standard deviation versus interpersonal standard deviation ratio (SDR). Experiment result shows that we can obtain feature measurements from barefoot impression images reliably. [ABSTRACT FROM AUTHOR]

A shadow detection scheme based on photometric invariant rgb color model is proposed. We firstly study the photometric invariance of rgb color model and deduce some important property. The algorithm combines the cues of moving cast shadow on brightness and chromaticity successively to detect candidate shadow regions in rgb color space; finally, a post-processing by exploiting region-based geometry information to exclude pseudo shadow segments. Results are presented for several video sequences representing a variety of illumination conditions and ground materials when the shadows are cast on different surface types. The results show our approach robust to widely different background and illuminations. [ABSTRACT FROM AUTHOR]

Searching and enhancement of shadow area in the satellite imagery is one of growing interest because of new possible needs of application in this field. This paper proposes an algorithm to search the shadow areas caused by buildings which are very common in satellite imagery of urban area in Korea. Binarization using histogram and threshold has demerits to have scattered small shadow areas which should be ignored for some applications. The proposed searching algorithm uses the fast Fourier transform and computes correlation in frequency domain. We search the threshold for correlation which is appropriate to obtain the shadow areas which do not include the scattered small dark areas. Experimental results show this method is valid to extract shadow areas from the satellite imagery. [ABSTRACT FROM AUTHOR]

Based on theoretical analysis of nonlinear feature extraction, we proposed a new method called a dynamic minimum subband spectral subtraction (DMSSS) and discussed its effects to the results of speech recognition. We illustrate the process of removing corrupted components by subtracting the estimated dynamic minimum of short-time spectra. Experimental results show the proposed method is stable and yield a good performance in ASR under noisy environments. If combined with peak isolation method, DMSSS can improve the recognition performance significantly. [ABSTRACT FROM AUTHOR]

A novel local feature detection method is presented for mobile robot's visual simultaneous localization and map building (v-SLAM). Camera-based visual localization can handle complicated problems, such as kidnapping and shadowing, which come with other type of sensors. Fundamental requirement of robust self-localization is robust key-point extraction under affine transform and illumination change. Especially, localization under low light environment is crucial for the purpose of guidance and navigation. This paper presents an efficient local feature extraction method under low light environment. A more efficient local feature detector and a compensation scheme of noise due to the low contrast images are proposed. The propose scene recognition method is robust against scale, rotation, and noise in the local feature space. We adopt the framework of scale-invariant feature transform (SIFT), where the difference of Gaussian (DoG)-based scale-invariant feature detection module is replaced by the difference of wavelet (DoW). [ABSTRACT FROM AUTHOR]

In this paper, we propose a music information retrieval (MIR) system. In the real mobile environment, a query music signal is captured by a cellular phone. A major problem in this environment is distortions contained in the features of the query sound due to the mobile network and environmental noise. In order to alleviate these noises, a signal subspace noise reduction algorithm is applied. Then a robust feature extraction method called Multi-Feature Clustering (MFC) combined with SFS feature optimization is implemented to improve and stabilize the system performance. The proposed system has been tested with using cellular phones in the real world and it shows about 65% of average retrieving success rate. [ABSTRACT FROM AUTHOR]

A region-based fuzzy shock filter with anisotropic diffusion is presented for image noise removal and edge sharpening. An image is divided into three-type different regions according to image features. For different regions, a binary shock-type backward diffusion or a fuzzy backward diffusion is performed in the gradient direction to the isophote line, incorporating a forward diffusion in the tangent direction. Gaussian smoothing to the second normal derivative results in a robust process against noise. Experiments on real images show that this method produces better visual results of the enhanced images than some related equations. [ABSTRACT FROM AUTHOR]

This paper describes the 3D reconstruction of planar objects using parallel lines from single panoramic image. Determination of non-vertical lines is depends on position of vanishing points with two lines on panoramic image. The vertical 3D line is projected as radial line on panoramic image and horizontal line is projected as curve or arc in panoramic image. Two parallel vertical lines are converged on center point in calibrated panoramic image. On the contrary, two parallel horizontal lines have the pair of vanishing points on the circle at infinity in panoramic image. We reconstruct the planar objects with parallel lines using the vanishing points and properties of parallelism in lines. Finally, we analysis and present the results of 3D reconstruction of planar objects by synthetic or real image. [ABSTRACT FROM AUTHOR]

The recognition of relatively big and rarely movable objects such as refrigerators and air conditioners, etc. is necessary because these objects can be crucial global features for Simultaneous Localization and Map building(SLAM) in indoor environment. In this paper, we propose a novel method to recognize these big objects using a sequence of 3D scenes. The particles which represent an object to be recognized are scattered into the 3D scene captured from an environment and then the probability of each particle is calculated by matching the 3D lines of the object model with them of the environment. Based on the probabilities and the degree of convergence of the particles, the object in the environment can be recognized and the position of the object can also be estimated. The experimental results show the feasibility of the suggested method based on particle filtering and its application to SLAM problems. [ABSTRACT FROM AUTHOR]

This article describes a fast and stable camera tracking method aimed for real-time augmented reality applications. From the feature tracking of a known marker on a single frame, we estimate the camera position and translation parameters. The entire pose estimation process is linear and initial estimates are not required. As an experimental setup, we implemented a video augmentation system to replace detected markers with virtual 3D graphical objects. Experimental results showed that the proposed camera tracking method is robust and fast applicable to interactive augmented reality applications. [ABSTRACT FROM AUTHOR]

An improved Genetic Algorithms (GA) for parameters estimation of multi-sine signals (PEMS) is proposed. The strategies of self-adaptive elite criterion, two-points crossover and cataclysmic mutation are employed in this algorithm to improve the performance of GA. For simplifying the computation, a complicated operating process is converted into several simple processes. A model of PEMS is also built which is conveniently applied to GA. Simulation results show that the proposed method is effective and superior to the least -mean squares (LMS) method. [ABSTRACT FROM AUTHOR]

This paper addresses the issue of detection and parameter estimation of multicomponent Polynomial Phase Signals (mc-PPS's) embedded in noise based on high-order ambiguity function (HAF). We first show how existing PHAF-based techniques (PHAF—product HAF) are inadequate mainly in providing reliable detection for mc-PPS. The main contribution of this paper is that we present a novel parameter estimation method. Firstly, given a set of time delay, it gives rise to a set of estimates of phase coefficients based on HAF. Then it produces a final estimate of phase coefficient by means of voting. The new method improves the probability of detection and estimation accuracy while avoiding the issue of threshold selection. Computer simulations are carried out to illustrate the advantage of the proposed method over existing techniques. [ABSTRACT FROM AUTHOR]

On-line signature verification is one of the most accepted means for personal verification. This paper proposes an on-line signature verification method based on Wavelet Transform (WT). Firstly, the method uses wavelet transform to exact characteristic points of 3-axis force and 2-dimension coordinate of signature obtained by the F-Tablet. And then it builds 5-dimension feature sequences and dynamically creates multi-templates using clustering. Finally, after the fusion of the above-mentioned 5-dimension feature sequences, whether the signature is genuine or not is decided by majority voting scheme. Experimenting on a signature database acquired by F-Tablet, the performance evaluation in even EER (Equal Error Rate) was improved to 2.83%. The experimental results show that the method not only reduces the amount of data to be stored, but also minimizes the duration of the whole authentication processing and increases the efficiency of signature verification. [ABSTRACT FROM AUTHOR]

In this paper, a novel scheme for fast object segmentation and tracking in MPEG-2 compressed domain is proposed. The video object is finally extracted after steps of motion detection, vector-based watershed segmentation, fusing operation, and finally edge correcting and morphologic post-processing. The tracking algorithm is fast and simple. All the processes are mainly implemented in compressed domain without the need of fully decoding of compressed stream. The information of motion vectors and DCT coefficients used in the algorithm are directly extracted from the compressed stream. Experimental results reveal that the proposed algorithm can extract objects directly from compressed stream with accurate contours and the object tracking algorithm is also efficient. [ABSTRACT FROM AUTHOR]

A simple, low computational cost and robust segmentation method is proposed by means of having successful experiences of strokes extraction for handwritten Chinese character and taking into account the characteristics of signature verification. After segmented and feature extracted, each signature is represented by a series of 6-dimensions vectors. Then, the degree of similarity between the questioned sample and 4 genuine signature samples stored in the reference database is calculated using these vectors. At last, the similarity vectors are inputted into RBFNN Classifier to decide whether the question sample is a genuine sample or not. The promising results of experiments indicate the segmentation method is fitting for Chinese signature verification and the whole verification method distinguish forgeries from genuine signatures effectively. [ABSTRACT FROM AUTHOR]

The ability to detect a new fault class can be a useful feature for an intelligent fault classification and diagnosis system. In this paper, we adopt two novelty detection methods, the support vector data description (SVDD) and the Parzen density estimation, to represent known fault class samples, and to detect new fault class samples. The experiments on real multi-class bearing fault data show that the SVDD can give both high identification rates for the prescribed ‘unknown' fault samples and the known fault samples, which shows an advantage over the Parzen density estimation method in our experiments, via choosing the appropriate SVDD algorithm parameters. [ABSTRACT FROM AUTHOR]

This paper presents a method that automatically detects the insect's abnormal movements. In general, the ecological data are difficult for analysis due to complexity residing in the systems with the variables varying in non-stationary fashion. Therefore, we needs to efficient methods that are able to measure from various environmental conditions. In this paper the wavelet transform are introduced as an alternative tool for extracting local and global information out of complex ecological data. And we discuss the method that is applicable to various relative fields. [ABSTRACT FROM AUTHOR]

In this paper, we proposed a deinterlacing algorithm using neural networks for conversion of interlaced videos to progressive videos. The proposed method uses multiple fields: a previous field, a current field, and a next field. Since the proposed algorithm uses multiple fields, the neural network is able to take into account the motion pattern which might exists in adjacent fields. Experimental results demonstrate that the proposed algorithm provides better performances than existing neural network deinterlacing algorithms that uses a single field. [ABSTRACT FROM AUTHOR]

This paper proposes two novel palmprint textural features, orientationCode and diffCode, and investigates the fusion of these features at score level for personal recognition. The orientationCode and diffCode are first defined using four directional templates and the differential operation, respectively. And then the matching score are computed to measure the similarity of the features. Finally, several fusion strategies are investigated for the matching scores of orientationCode and diffCode. Experimental results show that the orientationCode and diffCode can describe a palmprint effectively and the Sum, Product and Fisher's Linear Discriminant (FLD) fusion strategies can greatly improve the accuracy of palmprint authentication. [ABSTRACT FROM AUTHOR]

Video traffic predicting and modeling are very important for compressed video transmission. The traditional method delineates the process by a rigid model with several parameters, which are difficult to estimate. In this paper, the MPEG-4 VBR (Variable Bit Rate) video traffic is modeled by the ANFIS (Adaptive Neuro-Fuzzy Inference System). Then, it is applied to modeling and predicting the MPEG-4 VBR video traffic. Simulations show the GoP (Group of Pictures) loss probabilities in actual video traffic are very close to those in ANFIS modeling traffic at the same experimental conditions and the prediction errors (1/SNR) are very small. [ABSTRACT FROM AUTHOR]

In this paper, we describe an approach to inducing an expressive music performance model from a set of audio recordings of XVIII century bassoon pieces. We use a melodic transcription system which extracts a set of acoustic features from the recordings producing a melodic representation of the expressive performance played by the musician. We apply a machine learning techniques to this representation in order to induce a model of expressive performance. We use the model for both understanding and generating expressive music performances. [ABSTRACT FROM AUTHOR]

This paper presents a method for intelligent shape analysis of the hippocampus in a human brain using multi-sensory interface. To analyze the shape difference between two groups of the hippocampus, initially we extract quantitative shape features from input images, and then perform statistical shape analysis using parametric representation and Support Vector Machines (SVMs) learning algorithm. Results suggest that the presented shape representation and a polynomial kernel based SVMs algorithm can effectively discriminate between normal controls and epilepsy patients. To provide a more immersive and realistic environment in analysis, we combined a stereoscopic display and a 6-DOF force-feedback haptic device. The presented multi-sensory environment improves space and depth perception, and provides users with sense of touch feedback while making it easier to manipulate 3D objects. [ABSTRACT FROM AUTHOR]

H.264 allows motion estimation performing on multiple reference frames and seven modes. This new feature improves the prediction accuracy of inter-coding blocks significantly, but it is extremely computational intensive because the complexity of multi-frame motion estimation is quickly increased with the number of used reference frames. Moreover, the distortion gain caused by each reference frame in various modes is correlated, therefore it is not efficient to scan all the candidate frames in all seven modes. In this paper, a novel inherit-based adaptive frame selection method is proposed to reduce the complexity of the multi-frame motion estimation process. A new reference list for ME (Motion Estimation) of low level mode is constructed adaptively according to the results of ME of up level mode. Simulation results show that the proposed method can save about 15% to 50% computations while getting almost the same Rate-Distortion performance as the full scan. [ABSTRACT FROM AUTHOR]

A technology for automatic evaluation of images from protein crystallization trials is presented in this paper. In order to minimize the interference posed by the environmental factors, the droplet is segmented from the entire image first. The algorithm selects different features, which are derived from the pixels within the droplet, and obtains a 16-dimensional feature vector which will then be fed to the classifier to make a classification. Each image is classified into one of the following classes: "Clear", "Precipitate" and "Crystal". We have achieved an accuracy rate of 84.8% with our algorithm. [ABSTRACT FROM AUTHOR]

Though there have been proposed many magnification works in literatures, magnification in this paper is approached as reconstructing the geometric structures of the original high-resolution image. The structure tensor is able to estimate the orientation of both the edges and flow-like textures, which hence is much appropriate to magnification. Firstly, an edge-enhancing PDE and a corner-growing PDE are respectively proposed based on the structure tensor. Then, the two PDE's are combined into a novel one, which not only enables to enhance the edges and flow-like textures, but also to preserve the corner structures. Finally, the novel PDE is applied to image magnification. The method is simple, fast and robust to both the noise and the blocking-artifact. Experiment results demonstrate the effectiveness of our approach. [ABSTRACT FROM AUTHOR]

The control of crystal size distribution is critically important in crystallization process. Therefore the measurement of crystal size distribution attracts much attention. Image analysis is an advanced method developed recently for crystal size distribution estimation. A feasible image enhancement method is proposed for crystal identification in crystallization image by using histogram equalization and Laplacian mask algorithm sequentially. The experiments result indicates that the effect of the crystal image can be improved obviously, and the crystals can be identified more easily and exactly. [ABSTRACT FROM AUTHOR]

When developing intelligent recognition systems, our perception of patterns can be simulated using neural networks. An intelligent coin identification system that uses coin patterns for classification helps prevent confusion between different coins of similar physical dimensions. Currently, coin identification by machines relies on the assessment of the coin's physical parameters. In this paper, a rotation-invariant intelligent coin identification system (ICIS) is presented. ICIS uses a neural network and pattern averaging to recognize rotated coins at various degrees. Slot machines in Europe accept the new Turkish 1-Lira coin as a 2-Euro coin due to physical similarities. A 2-Euro coin is roughly worth 4 times the new Turkish 1-Lira. ICIS was implemented to identify the 2 EURO and 1 TL coins and the results were found to be encouraging. [ABSTRACT FROM AUTHOR]

For a corpus-based Mandarin text-to-speech system, the quality of synthesized speech is highly affected by the accuracy of unit boundaries. In this paper, we proposed a hybrid model method for automatic segmentation of Mandarin text-to-speech corpus. The boundaries of acoustic units are categorized into eleven phonetic groups. For a given phonetic group of boundaries, the proposed method selects an appropriate model from initial-final monophone-based HMM, semi-syllable monophone-based HMM and initial-final triphone-based HMM. The experimental results show that the hybrid model method can achieve better performance than the single model method, in terms of error rate and time shift of boundaries. [ABSTRACT FROM AUTHOR]

Though popularity and improvement of Internet with the explosive proliferation of multimedia contents bring us the era of digital information, the unexpected popularity of the Internet brings us its own dark side. Everyday young children are exposed to images that should not be delivered to them. In this paper, we propose an adult image rating system that properly classifies an image into one of multiple classes such as swimming suit, topless, all nude, sex image, and normal. The simulation results show that the proposed system successfully rates images into multiple classes with the rate of over 70%. [ABSTRACT FROM AUTHOR]

In this paper, a new shape representation method is proposed. We exploit our strategy in three steps. First we calculate the centroid and centroid distance of an image contour. Then based on polar coordinate system, the contour points are selected to construct graph, which is called Polar-Graph. The spectra of these graphs are finally organized as feature vectors for future clustering or retrieval. Our experiments show that the proposed representation is invariant to scale, translation and rotation, and is insensible to slight distortion and occlusion in some measure. [ABSTRACT FROM AUTHOR]

This paper proposes a new feature extraction method called grouping sampling reduction-based linear discriminant analysis. It solves the small sample size problem by using grouping sampling reduction to generate virtual samples with larger number and lower dimension than the original samples. The experiment result shows its efficiency and characteristic of high recognition rate. [ABSTRACT FROM AUTHOR]

As for the discriminant analysis on nonlinear manifold, a geodesic Gabriel graph based supervised manifold learning algorithm was proposed. Using geodesic distance to discover the intrinsic geometry of the manifold, the geodesic Gabriel graph was constructed to locate the key local regions where the local linear classifiers would be learned. The global nonlinear classifier was achieved by merging the multiple local classifiers applying the soft margin criterion, which assigned the optimal weight to each local classifier in an iterative way without any assumption of the distribution of the example data. The superiority of this algorithm is confirmed by experiments on synthesized data and face image databases. [ABSTRACT FROM AUTHOR]

Crying is the only communication way recently born babies have to express their needs. Several studies have shown that infant cry can be a valuable tool to determine the different infant's emotional, and physiological states. With the aim in usefully applying the crying information, in this paper we present the use of Fuzzy Support Vector Machines (FSVM) for two different infant cry recognition tasks. In the first one to identify pathologies, we classify Normal, Deaf, and Asphyxia infant cries. The second problem is about identifying Pain cries, Hunger cries and No-Pain-No-Hunger cries which are those that do not belong to any of the first two classes. Here we show that FSVM perform better than conventional SVM reaching a correct classification accuracy of up to 90%. [ABSTRACT FROM AUTHOR]

In this paper, we propose a multiscale feature extraction method of hand-vein patterns based on ridgelet transform and local interconnection structure neural networks. In order to restrain the noises and emphasize the hand-vein pattern in the image, we perform the multiscale self-adaptive enhancement transform based on the ridgelet transform to the hand-vein image. A neural network with local interconnection structure is designed to extract the features of the hand-vein patterns and deal with different size hand-vein patterns by using different receptive fields. By using the structural matching method we identify the hand-vein patterns. Our experimental results show that the proposed methods are superior to other methods and efficiently solve the problem of extracting features from the unclear images. But more experiments using a large database are need. [ABSTRACT FROM AUTHOR]

Electroencephalography (EEG) is a useful tool for observing brain activities. With the development of modern signal processing and statistics techniques, we have the ability to explore more information from EEG, much beyond classic trial-averaging approach. This paper discusses some characteristic methods for feature extraction and pattern classification on EEG data theoretically, and a combined spatial-temporal-frequency analysis strategy is proposed. After comparison of the linear function and support vector machine, reinforcement training is introduced to optimize a pre-designed linear classifier while at the same time restrain over-fitting of the classifier. Experiments performed on real EEG experimental data shows that it is effective in mining potential EEG patterns, and by which we achieved consistently results on all four submitted datasets of the BCI competition III. [ABSTRACT FROM AUTHOR]

The fall is a crucial problem in the elderly people's daily life, and the early detection of fall is very important to rescue the subjects and avoid the badly prognosis. In this paper, we use a wearable tri-axial accelerometer to capture the movement data of human body, and propose a novel fall detection method based on one-class support vector machine (SVM). The one-class SVM model is trained by the positive samples from the falls of younger volunteers and a dummy, and the outliers from the non-fall daily activities of younger and the elderly volunteers. The preliminary results show that this method can detect the falls effectively, and reduce the probability of being damaged in the experiments for the elderly people. [ABSTRACT FROM AUTHOR]

In this paper, we propose a new method of face detection using an adaptive skin color model and eye features. First, we detect skin color segments adaptively using a two-dimensional Gaussian model from the CrCb skin color space. On these skin color segments, shape analysis is performed to reduce false alarms. Then, eye feature points for face are extracted. The possible eye feature points are compared with normalized eye features obtained from the training data for verification. At this step, we use a modified Hausdorff distance. Experimental results are given to demonstrate our face detection approach in slanted face images and different lighting conditions. [ABSTRACT FROM AUTHOR]