Search

The www is a dynamic environment and it is difficult to capture the user preferences and interests without interfering with the normal activity of the user. Hence to improve user searching in the World Wide Web, we have proposed a personalized search system that supports user searches by learning about user preferences and by observing responses to prior search experiences aided by a new index called the User Conceptual Index (UCI). This paper models every user's search behavior as a User's Search Behavior (USB) Graph. The main focus of this paper is the analysis of the USB graph and the redesign of the UCI using the results arrived from the analysis. [ABSTRACT FROM AUTHOR]

In this paper we present a well designed method that makes use of edge information to extract textual blocks from gray scale document images. It aims at detecting textual regions on heavy noise infected newspaper images and separate them from graphical regions. The algorithm traces the feature points in different entities and then groups those edge points of textual regions. Finally feature based connected component merging was introduced to gather homogeneous textual regions together within the scope of its bounding rectangles. The proposed method can be used to locate text in-group of newspaper images with multiple page layouts. Initial results are encouraging, then they are experimented with considerable number of newspaper images with different layout structures and promising results were obtained. This finds its major application in digital libraries for OCR where information can be of different quality depending on the age of the scanned paper. [ABSTRACT FROM AUTHOR]

In this paper, we present a HMM based system that is used to recognize ultra low resolution text such as those frequently embedded in images available on the web. We propose a system that takes specifically the challenges of recognizing text in ultra low resolution images into account. In addition to this, we show in this paper that word models can be advantageously built connecting together sub-HMM-character models and inter-character state. Finally we report on the promising performance of the system using HMM topologies which have been improved to take into account the presupposed minimum length of each character. [ABSTRACT FROM AUTHOR]

An automated approach for computation of the frontal plane QRS vector and an important observation of its clinical significance is described in this paper. Frontal plane QRS vector is computed from the six frontal plane leads ( Standard leads I, II, III , AVR, AVL and AVF). The R-R interval of each ECG wave is detected by square derivative technique. The baseline or isoelectric level of every ECG wave is determined. After that the net positive or net negative deflection (NQD) of QRS complex is detected. Net positive or net negative deflection in any lead is obtained by subtracting the smaller deflection (+ve or -ve) from the larger deflection (-ve or +ve). An algorithm is developed for computation of the exact angle,amplitude and direction of the frontal plane QRS vector from maximum and minimum NQD. In the present work, the PTB diagnostic ECG database of normal and Myocardial Infarction (MI) subjects is used for computation of the QRS vector. An interesting clinical observation that, the rotation of QRS axis for MI data may significantly detect the region of the infarcted cardiac wall, is reported in this paper. [ABSTRACT FROM AUTHOR]

Automatic tracking of references involves aggregating and synthesizing references through World Wide Web, thereby introducing greater efficiency and granularity to the task of finding publication information. This paper discusses the design and implementation of crawler-based reference tracking system, which has the advantage of online reference filtering. The system automatically analyses the semantic relevance of the reference article by harvesting keywords and their meanings, from title and abstract of the respective article. Indirectly this attempts to improve the performance of the reference database by reducing the articles that are actually being downloaded thereby improving the performance of the system. The number of levels for recursive downloads of reference articles are specified by the user. According to user's interest the system tracks up the references required for the understanding of the seed article, stores them in the databases and projects the information by threshold based view filtering. [ABSTRACT FROM AUTHOR]

This paper presents a framework for using a case-based reasoning system for stock analysis in financial market. The unique aspect of this paper is the use of a hierarchical structure for case representation. The system further incorporates a multi-criteria decision-making algorithm which furnishes the most suitable solution with respect to the current market scenario. Two important aspects of financial market are addressed in this paper: stock evaluation and investment planning. CBR and multi-criteria when used in conjunction offer an effective tool for evaluating goodness of a particular stock based on certain factors. The system also suggests a suitable investment plan based on the current assets of a particular investor. Stock evaluation maps to a flat case structure, but investment planning offers a scenario more suited for structuring the case into successive detailed layers of information related to different facets. This naturally leads to a hierarchical case structure. [ABSTRACT FROM AUTHOR]

This paper presents a computer vision guidance system for agricultural vehicles. This system is based on a segmentation algorithm that uses an optimum threshold function in terms of minimum quadratic value over a discriminant based on the Fisher lineal discriminant. This system has achieved not only very interesting results in the sense of segmentation, but it has also guided successfully a vehicle in a real world environment. [ABSTRACT FROM AUTHOR]

Transmission of images and video over unreliable channels produces visual artifacts due to either loss or corruption of bit streams. Error concealment schemes aim at reducing such degradation and improving the visual quality. Error concealment schemes usually conceal errors of a specific type. This points towards the possibility of combining more than one scheme, for efficient error concealment and better visual effects. In this paper, we propose a scheme for error concealment using a combination of watermarking and Projection Onto Convex Sets (POCS). Watermarking an image using the information derived from the host data itself conceals errors initially and those which are prominent even after this stage are concealed through POCS. The proposed method can be used for concealing errors in images and intra coded video by preserving average values through watermarking and edge information through POCS. Simulation results of POCS, watermarking and watermarking with POCS are given in the paper to show the advantage of combining both the methods. [ABSTRACT FROM AUTHOR]

Applications in various domains often lead to very large and frequently high-dimensional data. Successful algorithms must avoid the curse of dimensionality but at the same time should be computationally efficient. Finding useful patterns in large datasets has attracted considerable interest recently. The primary goal of the paper is to implement an efficient Hybrid Tree based clustering method based on CF-Tree and KD-Tree, and combine the clustering methods with KNN-Classification. The implementation of the algorithm involves many issues like good accuracy, less space and less time. We will evaluate the time and space efficiency, data input order sensitivity, and clustering quality through several experiments. [ABSTRACT FROM AUTHOR]

The discrete wavelet transforms (DWTs) applied temporally under motion compensation (i.e. Motion Compensation Temporal Filtering (MCTF)) has recently become a very powerful tool in scalable video compression, especially when implemented through lifting. The major bottleneck for speed of the encoder is the computational complexity of the bidirectional motion estimation in MCTF. This paper proposes a novel predictive technique to reduce the computational complexity of MCTF. In the proposed technique the temporal filtering is done without motion compensation. The resultant high frequency frames are used to predict the blocks under motion. Motion estimation is carried out only for the predicted blocks under motion. This significantly reduces the number of blocks that undergoes motion estimation and hence the computationally complexity of MCTF is reduced by 44% to 92% over variety of standard test sequences without compromising the quality of the decoded video. The proposed algorithm is implemented in MC-EZBC, a 3D-subband scalable video coding system. [ABSTRACT FROM AUTHOR]

Robust tracking of objects in video is a key challenge in computer vision with applications in automated surveillance, video indexing, human-computer-interaction, gesture recognition, traffic monitoring, etc. Many algorithms have been developed for tracking an object in controlled environments. However, they are susceptible to failure when the challenge is to track multiple objects that undergo appearance change to due to factors such as variation in illumination and object pose. In this paper we present a tracker based on Bayesian estimation, which is relatively robust to object appearance change, and can track multiple targets simultaneously in real time. The object model for computing the likelihood function is incrementally updated and uses background-foreground segmentation information to ameliorate the problem of drift associated with object model update schemes. We demonstrate the efficacy of the proposed method by tracking objects in image sequences from the CAVIAR dataset. [ABSTRACT FROM AUTHOR]

In case of monitoring road traffic,the image based monitoring system is more useful than any other system such as GPS or loop detector because it can give the whole picture of the two-dimensional traffic situation. The idea of this paper is that the quad-tree scheme segments MBR following from the background subtraction process. Then the segmented and detected vehicle regions, ROIs, are tracked by SIFT algorithm. Our method succeeded detecting and tracking multiple moving vehicles accurately in sequence frame. The proposed method is very useful for the video based applications such as automatic traffic monitoring system. [ABSTRACT FROM AUTHOR]

This paper presents an architecture that enables the recognizer to learn incrementally and, thereby adapt to document image collections for performance improvement. We argue that the recognition scheme for a book could be considerably different from that designed for isolated pages. We employ learning procedures to capture the relevant information available online, and feed it back to update the knowledge of the system. Experimental results show the effectiveness of our design for improving the performance on-the-fly. [ABSTRACT FROM AUTHOR]

Named Entity Recognition (NER) has an important role in almost all Natural Language Processing (NLP) application areas including information retrieval, machine translation, question-answering system, automatic summarization etc. This paper reports about the development of a statistical Hidden Markov Model (HMM) based NER system. The system is initially developed for Bengali using a tagged Bengali news corpus, developed from the archive of a leading Bengali newspaper available in the web. The system is trained with a training corpus of 150,000 wordforms, initially tagged with a HMM based part of speech (POS) tagger. Evaluation results of the 10-fold cross validation test yield an average Recall, Precision and F-Score values of 90.2%, 79.48% and 84.5%, respectively. This HMM based NER system is then trained and tested on the Hindi data to show its effectiveness towards the language independent abilities. Experimental results of the 10-fold cross validation test has demonstrated the average Recall, Precision and F-Score values of 82.5%, 74.6% and 78.35%, respectively with 27,151 Hindi wordforms. [ABSTRACT FROM AUTHOR]

The document images that are fed into an Optical Character Recognition system, might be skewed. This could be due to improper feeding of the document into the scanner or may be due to a faulty scanner. In this paper, we propose a skew detection and correction method for document images. We make use of the inherent randomness in the Horizontal Projection profiles of a text block image, as the skew of the image varies. The proposed algorithm has proved to be very robust and time efficient. The entire process takes less than a second on a 2.4 GHz Pentium IV PC. [ABSTRACT FROM AUTHOR]

Voice transformation involves modifying the source speaker voice to target speaker voice. Voice characteristics of a speaker depends on the shape of the glottal pulse (source characteristics), shape of the vocal tract system (system characteristics) and the long term features (prosody or supra-segmental) of the speech signal produced by the speaker. In this paper we proposed the mapping functions to transform the vocal tract characteristics and intonation characteristics from source speaker to target speaker. Mapping functions are developed by the features extracted from syllable level. The shape of the vocal tract system is characterized by linear prediction coefficients, and the mapping function is realized by a five layer feedforward neural network. Mapping of the intonation characteristics (pitch contour) is provided by associating the code books derived from the pitch contours of the source and target speakers. The proposed mapping functions are used in voice transformation task. The target speaker's speech is synthesized and evaluated using listening tests. The results of the listening tests indicate that the proposed voice transformation provides better mapping of the voice characteristics compared to the earlier method proposed by the author. The original and the synthesized speech signals obtained using mapping functions are available for listening at http://shilloi.iitg.ernet.in/~ksrao/result.html [ABSTRACT FROM AUTHOR]

During document scanning, skew is inevitably introduced into the incoming document image. Presence of additional modified characters, which get plugged in as extensions and remain as disjointed protrusions of a main character is really challenging in estimating inclination in skewed documents made up of texts in south Indian languages (Kannada, Telugu, Tamil and Malayalam). In this paper, we present a novel script independent (for south Indian) skew estimation technique based on Gaussian Mixture Models (GMM). The Expectation-Maximization (EM) algorithm is used to learn the mixture of Gaussians. Subsequently the cluster means are subjected to moments to estimate the skew angle. Experiments on printed and handwritten documents corrupted by noise is done. Our method shows significantly improved performance as compared to other existing methods. [ABSTRACT FROM AUTHOR]

Language Identification (LID) refers to the task of identifying an unknown language from the test utterances. In this paper, a new feature set, viz.,T-MFCC by amalgamating Teager Energy Operator (TEO) and well-known Mel frequency cepstral coefficients (MFCC) is developed. The effectiveness of the newly derived feature set is demonstrated for identifying perceptually similar Indian languages such as Hindi and Urdu. The modified structure of polynomial classifier of 2nd and 3rd order approximation has been used for the LID problem. The results have been compared with state-of-the art feature set, viz.,MFCC and found to be effective (an average jump 21.66%) in majority of the cases. This may be due to the fact that the T-MFCC represents the combined effect of airflow properties in the vocal tract (which are known to be language and speaker dependent) and human perception process for hearing. [ABSTRACT FROM AUTHOR]

Performance of Language Identification (LID) System using Gaussian Mixture Models (GMM) is limited by the convergence of Expectation Maximization (EM) algorithm to local maxima. In this paper an LID system is described using Gaussian Mixture Models for the extracted features which are then trained using Split and Merge Expectation Maximization Algorithm that improves the global convergence of EM algorithm. It improves the learning of mixture models which in turn gives better LID performance. A maximum likelihood classifier is used for classification or identifying a language. The superiority of the proposed method is tested for four languages [ABSTRACT FROM AUTHOR]

Level crossing based sampling might be used as an alternative to Nyquist theory based sampling of a signal. Level crossing based approach take advantage of statistical properties of the signal, providing cues to efficient nonuniform sampling. This paper presents new threshold level allocation schemes for level crossing based nonuniform sampling. Intuitively, it is more reasonable if the information rich regions of the signal are sampled finer and those with sparse information are sampled coarser. To achieve this objective, we proposed non-linear quantization functions which dynamically assign the number of quantization levels depending on the importance of the given amplitude range. Various aspects of proposed techniques are discussed and experimentally validated. Its efficacy is investigated by comparison with Nyquist based sampling. [ABSTRACT FROM AUTHOR]

A central and long-standing objective of cellular physiology has been to understand the metabolic capabilities of living cells. In this paper, we perform optimization of metabolic networks in E.Coli under both flux and energy balance constraints. The impact of the energy balance is investigated on the behavior of metabolic networks. Different from the existing work, the preliminary results showed that the impact of imposing energy balance constraints on the biological metabolic systems is very limited. [ABSTRACT FROM AUTHOR]

Conflict graph is used as the major tool in various algorithms [14] - [18] for solving the phylogenetic network problem. The over all time complexity of these algorithms mainly depends on the construction of the conflict graph. In this paper, we present a parallel algorithm for building a conflict graph. Given a set of n binary sequences, each of size m, our algorithm is mapped on a triangular array in O(n) time using O(m2) processors. [ABSTRACT FROM AUTHOR]

Semantic frameworks can be used to improve the accuracy and expressiveness of natural language processing for the purpose of extracting meaning from text documents. Such a framework represents knowledge using semantic networks and can be generated using information mined from text documents. The key issue however is to identify relevant concepts and their inter-relationships. In this paper, we have presented a scheme for semantic integration of information extracted from text documents. The extraction principle is based on linguistic and semantic analysis of text. Entities and relations are extracted using Natural Language Processing techniques. A method for collating information extracted from multiple sources to generate the semantic net is also presented. The efficacy of the proposed semantic framework is established through experiments carried out for visualizing information embedded in biomedical texts extracted from PubMed database. [ABSTRACT FROM AUTHOR]

This paper proposes a scheme for designing a classier along with fuzzy set selection. It detects discontinuities in the domain of input and prunes the corresponding fuzzy sets using a neuro-fuzzy architecture. This reduces the size of the network and so the number of rules. The network is trained in three phases. In the rst phase, the network learns the important fuzzy sets. In the subsequent phases, the network is pruned to produce optimal rule set by pruning conicting and less significant rules. We use a four-layered feed-forward network and error back propagation learning. The second layer of the network learns a modulator function for each input fuzzy set that identies the unnecessary fuzzy sets. In this paper, we also introduced the notion of utility factors for fuzzy rules. Rules with small utility factors are less signicant or less used rules and can be eliminated. They are learned and detected by the third layer. After training in phase 3 in its reduced structure, the system retains almost the same level of performance. The proposed system has been tested on synthetic data set and found to perform well. [ABSTRACT FROM AUTHOR]

The previously proposed cunning ant system (cAS), a variant of the ACO algorithm, worked well on the TSP and the results showed that the cAS could be one of the most promising ACO algorithms. In this paper, we apply cAS to solving QAP. We focus our main attention on the effects of applying local search and parallelization of the cAS. Results show promising performance of cAS on QAP. [ABSTRACT FROM AUTHOR]

Difficult knapsack problems are problems that are expressly designed to be difficult. In this paper, enhanced Quantum Evolutionary Algorithms are designed and their application is presented for the solution of the DKPs. The algorithms are general enough and can be used with advantage in other subset selection problems. [ABSTRACT FROM AUTHOR]

Boundary identification in medical images plays a crucial role in helping physicians in patient diagnosis. Manual identification of object boundaries is a time-consuming task and is subject to operator variability. Fully automatic procedures are still far from satisfactory in most real situations. In this paper, we propose a boundary identification method based on multiscale technique. Experimental results have shown that the proposed method provides superior performance in medical image segmentation. [ABSTRACT FROM AUTHOR]

In this paper, we compare several detection algorithms that are based on spectral matched (subspace) filters. Nonlinear (kernel) versions of these spectral matched (subspace) detectors are also discussed and their performance is compared with the linear versions. Several well-known matched detectors, such as matched subspace detector, orthogonal subspace detector, spectral matched filter and adaptive subspace detector (adaptive cosine estimator) are extended to their corresponding kernel versions by using the idea of kernel-based learning theory. In kernel-based detection algorithms the data is implicitly mapped into a high dimensional kernel feature space by a nonlinear mapping which is associated with a kernel function. The detection algorithm is then derived in the feature space which is kernelized in terms of the kernel functions in order to avoid explicit computation in the high dimensional feature space. Experimental results based on real hyperspectral imagery show that the kernel versions of these detectors outperform the conventional linear detectors. [ABSTRACT FROM AUTHOR]

In this paper, we address the problem of quick retrieval of online signatures. The proposed methodology retrieves signatures in the database for a given query signature according to the decreasing order of their spatial similarity with the query. Similarity is computed based on orientations of corresponding edges drawn in between sampled points of the signatures. We retrieve the best hypotheses in a simple yet efficient way to speed up the subsequent recognition stage. The runtime of the signature recognition process is reduced, because the scanning of the entire database is narrowed down to contrasting the query with a few top retrieved hypotheses. The experimentation conducted on a large MCYT_signature database [1] has shown promising results. [ABSTRACT FROM AUTHOR]

This paper describes a system for isolated Kannada handwritten numerals recognition using image fusion method. Several digital images corresponding to each handwritten numeral are fused to generate patterns, which are stored in 8x8 matrices, irrespective of the size of images. The numerals to be recognized are matched using nearest neighbor classifier with each pattern and the best match pattern is considered as the recognized numeral.The experimental results show accuracy of 96.2% for 500 images, representing the portion of trained data, with the system being trained for 1000 images. The recognition result of 91% was obtained for 250 test numerals other than the trained images. Further to test the performance of the proposed scheme 4-fold cross validation has been carried out yielding an accuracy of 89%. [ABSTRACT FROM AUTHOR]

During the execution of Web Service Composition, if one component service fails or becomes overloaded not to be accessed, a failure recovery mechanism is needed to ensure that the running process is not interrupted and the failed service can be replaced quickly and efficiently. Recent researches on this problem have some disadvantages. They don't consider the influence of the number of service candidates or the connection state of the overlay network, so the algorithms are easily disabled. In this paper, we present an adaptive algorithm to find replacement path locally by virtue of the old path during dynamic service composition. We go backward along the execution path to find the branch node, and then construct the sub-graph by the predefined rules. Finally we choose the best path with the highest total utility to replace the failed one. The test's result shows the algorithm performs very well in the vigor-ousness and availability to dynamic adaptation. [ABSTRACT FROM AUTHOR]

Leaders clustering method is a fast one and can be used to derive prototypes called leaders from a large training set which can be used in designing a classifier. Recently nearest leader based classifier is shown to be a faster version of the nearest neighbor classifier, but its performance can be a degraded one since the density information present in the training set is lost while deriving the prototypes. In this paper we present a generalized weighted k-nearest leader based classifier which is a faster one and also an on-par classifier with the k-nearest neighbor classifier. The method is to find the relative importance of each prototype which is called its weight and to use them in the classification. The design phase is extended to eliminate some of the noisy prototypes to enhance the performance of the classifier. The method is empirically verified using some standard data sets and a comparison is drawn with some of the earlier related methods. [ABSTRACT FROM AUTHOR]

Support vector machine (SVM) which was originally designed for binary classification has achieved superior performance in various classification problems. In order to extend it to multiclass classification, one popular approach is to consider the problem as a collection of binary classification problems. Majority voting or winner-takes-all is then applied to combine those outputs, but it often causes problems to consider tie-breaks and tune the weights of individual classifiers. This paper presents two novel ensemble approaches: probabilistic ordering of one-vs-rest (OVR) SVMs with naïve Bayes classifier and multiple decision templates of OVR SVMs. Experiments with multiclass datasets have shown the usefulness of the ensemble methods. [ABSTRACT FROM AUTHOR]

This paper presents a novel technique to jointly represent the shape and motion of video objects for the purpose of content based video retrieval (CBVR). It enables to retrieve similar objects undergoing similar motion patterns, that are not captured only using motion trajectory or shape descriptors. In our approach, both shape and motion information are integrated in a unified spatio-temporal representation. Curvature scale space theory proposed by Mokhtarian is extended (in 3D) to represent shape as well as motion trajectory of video objects. A sequence of 2D contours are taken as input and convolved with a 2D Gaussian. The zero crossings are found out from the curvature of evolved surfaces, which form the 3D CSS surface. The peaks from the 3D CSS surface form the features for joint spatio-temporal representation of video objects. Experiments are carried out on CBVR and results show good performance of the algorithm. [ABSTRACT FROM AUTHOR]

This paper presents an efficient technique for accurate detection of iris boundary, which is an important issue for any iris-based biometric identification system. Our proposed technique follows scaling, histogram equalization, edge detection and finally removal of unnecessary edges present in the eye image. Scaling and removing unnecessary edges enables us to reduce the search space for iris boundary. Experimental results show that with our approach it is possible to detect iris boundary as much as 98% of the eye images in CASIA database accurately and it needs only 25% time compared to the existing approaches. [ABSTRACT FROM AUTHOR]

In this paper, we present an approach for human fall detection, which has important applications in the field of safety and security. The proposed approach consists of two parts: object detection and the use of a fall model. We use an adaptive background subtraction method to detect a moving object and mark it with its minimum-bounding box. The fall model uses a set of extracted features to analyze, detect and confirm a fall. We implement a two-state finite state machine (FSM) to continuously monitor people and their activities. Experimental results show that our method can detect most of the possible types of single human falls quite accurately. [ABSTRACT FROM AUTHOR]

Temporal registration refers to the methods used to align time varying sample sets with respect to each other. While reconstruction from a single sample set may generate aliasing, registration of multiple sample sets increases the effective sampling rate and therefore helps alleviate the problems created by low acquisition rates. However, since registration is mostly computed as an iterative best estimate, any error in registration translates directly into an increase in reconstruction error. In this paper we present a confidence measure based on local and global temporal registration errors, computed between sample sets, to determine if a given set of samples is suitable for inclusion in the reconstruction of a higher resolution temporal dataset. We also discuss implications of the non-uniform sampling theorem on the proposed confidence measure. Experimental results with real and synthetic data are provided to validate the proposed confidence measure. [ABSTRACT FROM AUTHOR]

Gait along with body structure has been recognized as a potential biometric feature for identifying human beings. The spatial and temporal shape of motion of an individual is usually the same for all gait cycles and is considered to be unique to that individual. In this paper we introduce a Temporal Slice based approach for gait recognition. Temporal Slices are a set of two-dimensional images extracted along the time dimension of an image volume. They encode a rich set of visual patterns for similarity measure and have been widely used for motion detection. We show that the features obtained from tensor histogram of these temporal slices can be efficiently used as gait features for recognition of human beings. [ABSTRACT FROM AUTHOR]

The paper focuses on an evaluation of particle filter based facial feature tracker. Particle filter is a successful tool in the non-linear and the non-Gaussian estimation problems. We developed a particle filter based facial points tracker with a simple observation model based on sum-of-squared differences (SSD) between the intensities. Multistate face component model is used to estimate the occluded feature points. The important distances are calculated from tracked points. Two kinds of classification schemes are considered, the hidden Markov model (HMM) as sequence based recognizer and support vector machine (SVM) as frame based recognizer. A comparative study is shown in the classification of five basic expressions, i.e., anger, sadness, happiness, surprise and disgust. The tests are conducted on Cohn-Kanade and MMI face expression databases. [ABSTRACT FROM AUTHOR]

This paper presents a novel age transformation algorithm to handle the challenge of facial aging in face recognition. The proposed algorithm registers the gallery and probe face images in polar coordinate domain and minimizes the variations in facial features caused due to aging. The efficacy of the proposed age transformation algorithm is validated using 2D log polar Gabor based face recognition algorithm on a face database that comprises of face images with large age progression. Experimental results show that the proposed algorithm significantly improves the verification and identification performance. [ABSTRACT FROM AUTHOR]

The locality preserving projection (LPP), known as Laplacianfaces, was recently proposed as a transformation technique of mapping which optimally preserves the neighborhood structure of the dataset. In this paper, an efficient method for face recognition called mixture-of-Laplacianfaces (or LPP mixture model) is proposed, which obtains several sets of Laplacianfaces through Expectation-Maximization (EM) learning of Gaussian Mixture Models (GMM). Experiments carried out by using this on ORL, FERET and COIL-20 indicate superior performance as compared with method based on Laplacianfaces and other contemporary subspace methods. [ABSTRACT FROM AUTHOR]

This paper presents a recognition system for isolated handwritten Bangla words, with a fixed lexicon, using a Hidden Markov Model (HMM). A stochastic search method, namely, Genetic Algorithm (GA) is used to train the HMM. The HMM is a left-right HMM. For feature extraction, the image boundary is traced both in the anticlockwise and clockwise directions and the significant changes in direction along the boundary are noted. Certain features defined on the basis of these changes are used in the proposed model. [ABSTRACT FROM AUTHOR]

Keywords characterize the topics discussed in a document. Extracting a small set of keywords from a single document is an important problem in text mining. We propose a hybrid structural and statistical approach to extract keywords. We represent the given document as an undirected graph, whose vertices are words in the document and the edges are labeled with a dissimilarity measure between two words, derived from the frequency of their co-occurrence in the document. We propose that central vertices in this graph are candidates as keywords. We model importance of a word in terms of its centrality in this graph. Using graph-theoretical notions of vertex centrality, we suggest several algorithms to extract keywords from the given document. We demonstrate the effectiveness of the proposed algorithms on real-life documents. [ABSTRACT FROM AUTHOR]

In this paper, we propose hybrid fusion of audio and explicit correlation features for speaker identity verification applications. Experiments were performed with the GMM based speaker models with a hybrid fusion technique involving late fusion of explicit cross-modal fusion features, with implicit eigen lip and audio MFCC features. An evaluation of the system performance with different gender specific datasets from controlled VidTIMIT data base and opportunistic UCBN database shows a significant performance improvement. [ABSTRACT FROM AUTHOR]

It is not a surprise that image processing is a growing research field. Vision in general and images in particular have always played an important and essential role in human life. Not only as a way to communicate, but also for commercial, scientific, industrial and military applications. Many techniques have been introduced and developed to deal with all the challenges involved with image processing. In this paper, we will focus on techniques that find their origin in fuzzy set theory and fuzzy logic. We will show the possibilities of fuzzy logic in applications such as image retrieval, morphology and noise reduction by discussing some examples. Combined with other state-of-the-art techniques they deliver a useful contribution to current research. [ABSTRACT FROM AUTHOR]

An important goal in data mining is to generate an abstraction of the data. Such an abstraction helps in reducing the time and space requirements of the overall decision making process. It is also important that the abstraction be generated from the data in small number of scans. In this paper we propose a novel scheme called Prefix-Suffix trees for compact storage of patterns in data mining, which forms an abstraction of the patterns, and which is generated from the data in a single scan. This abstraction takes less amount of space and hence forms a compact storage of patterns. Further, we propose a clustering algorithm based on this storage and prove experimentally that this type of storage reduces the space and time. This has been established by considering large data sets of handwritten numerals namely the OCR data, the MNIST data and the USPS data. The proposed algorithm is compared with other similar algorithms and the efficacy of our scheme is thus established. [ABSTRACT FROM AUTHOR]

Support Vector Clustering has gained reasonable attention from the researchers in exploratory data analysis due to firm theoretical foundation in statistical learning theory. Hard Partitioning of the data set achieved by support vector clustering may not be acceptable in real world scenarios. Rough Support Vector Clustering is an extension of Support Vector Clustering to attain a soft partitioning of the data set. But the Quadratic Programming Problem involved in Rough Support Vector Clustering makes it computationally expensive to handle large datasets. In this paper, we propose Rough Core Vector Clustering algorithm which is a computationally efficient realization of Rough Support Vector Clustering. Here Rough Support Vector Clustering problem is formulated using an approximate Minimum Enclosing Ball problem and is solved using an approximate Minimum Enclosing Ball finding algorithm. Experiments done with several Large Multi class datasets such as Forest cover type, and other Multi class datasets taken from LIBSVM page shows that the proposed strategy is efficient, finds meaningful soft cluster abstractions which provide a superior generalization performance than the SVM classifier. [ABSTRACT FROM AUTHOR]

Identification of nonlinear static and dynamic systems plays a key role in many engineering applications including communication, control and instrumentation. Various adaptive models have been suggested in the literature using ANN and Fuzzy logic based structures. In this paper we employ an efficient and low complexity Functional Link ANN (FLANN) model for identifying such nonlinear systems using GA based learning of connective weights. In addition, pruning of connecting paths is also simultaneously carried out using GA to reduce the network architecture. Computer simulations on various static and dynamic systems indicate that there is more than 50% reduction in original model FLANN structure with almost equivalent identification performance. [ABSTRACT FROM AUTHOR]

In this paper, we compare performance of novel neural network based algorithm for Boolean factor analysis with several dimension reduction techniques as a tool for feature extraction. Compared are namely singular value decomposition, semi-discrete decomposition and non-negative matrix factorization algorithms, including some cluster analysis methods as well. Even if the mainly mentioned methods are linear, it is interesting to compare them with neural network based Boolean factor analysis, because they are well elaborated. Second reason for this is to show basic differences between Boolean and linear case. So called bars problem is used as the benchmark. Set of artificial signals generated as a Boolean sum of given number of bars is analyzed by these methods. Resulting images show that Boolean factor analysis is upmost suitable method for this kind of data. [ABSTRACT FROM AUTHOR]

This paper describes the design of fuzzy logic based sumo wrestling robot. The designed robot has a simple control algorithm and single fuzzy microcontroller is used in hardware implementation. The designed robot meets the specifications needed to compete in a sumo robot competition. The main difference of the designed system with earlier sumo robots is in control algorithms. A simple fuzzy logic controller is developed for detection and tracking of opponent in competition ring. Three infrared (IR) sharp sensors are used for target detection. Fuzzy microcontroller fuses the sensor data's and provides the necessary control signal to motors for heading robot toward the opponent. The fuzzy rules were optimized for the best results possible in software which are loaded in fuzzy controller. The implemented control algorithm shows better performance and executes the opponent detection algorithm in less time in comparison with conventional sumo robot algorithm. Design procedure and experimental results are presented to show the performance of the intelligent controller in designed system. [ABSTRACT FROM AUTHOR]