Discriminative subspace analysis is a popular approach for a variety of applications. There is a growing interest in subspace learning techniques for face recognition. Principal component analysis (PCA) and eigenfaces are two important subspace analysis methods have been widely applied in a variety of areas. However, the excessive dimension of data space often causes the curse of dimensionality dilemma, expensive computational cost, and sometimes the singularity problem. In this paper, a new supervised discriminative subspace analysis is presented by encoding face image as a high order general tensor. As face space can be considered as a nonlinear submanifold embedded in the tensor space, a decomposition method called Tucker tensor is used which can effectively decomposes this sparse space. The performance of the proposed method is compared with that of eigenface, Fisherface, tensor LPP, and ORO4×2 on ORL and Weizermann databases. Conducted experimental results show the superiority of the proposed method.

Web access prediction has attracted significant attention in recent years. Web prefetching and some personalization systems use prediction algorithms. Most current applications that predict the next user web page have an offline component that does the data preparation task and an online section that provides personalized content to the users based on their current navigational activities. In this paper we present an online prediction model that does not have an offline component and fit in the memory with good prediction accuracy. Our algorithm is based on LZ78 and LZW algorithms that are adapted for modeling the user navigation in web. Our model decreases computational complexities which is a serious problem in developing online prediction systems. A performance evaluation is presented using real web logs. This evaluation shows that our model needs much less memory than PPM family of algorithms with good prediction accuracy.

In this paper, we propose a method for selecting the symmetry axis of eyes region from two or more candidates. We propose a region-based deformable template matching from two new defined operations: intensity-based 2-clustering and edge shadowing. The results display the effectiveness of our method for extraction of eye, eyebrow and nose templates. The parameters of these templates can be used as feature vectors in low bit rate transmission. Evaluation of the proposed method on an Iranian database shows the accuracy of 99% for feature region extraction and 86% in average for feature template extraction.

This paper presents an efficient algorithm for adaptive binarization of degraded document images. Document binarization algorithms suffer from poor and variable contrast in document images. We propose a contrast independent binarization algorithm that does not require any parameter setting by user. Therefore, it can handle various types of degraded document images. The proposed algorithm involves two consecutive stages. At the first stage, independent of contrast between foreground and background, some parts of each character are extracted and in the second stage, the gray level of foreground and background are locally estimated. For each pixel, the average of estimated foreground and background gray levels is defined as threshold. After extensive experiments, the proposed binarization algorithm demonstrate superior performance against four well-know binarization algorithms on a set of degraded document images captured with camera.

One of the important requirements for operational planning of electrical utilities is the prediction of hourly load up to several days, known as Short Term Load Forecasting (STLF). Considering the effect of its accuracy on system security and also economical aspects, there is an on-going attention toward putting new approaches to the task. Recently, Neuro Fuzzy modeling has played a successful role in various applications over nonlinear time series prediction. This paper presents a neuro-fuzzy model for the application of short-term load forecasting. This model is identified through Locally Liner Model Tree (LoLiMoT) learning algorithm. The model is compared to a multilayer perceptron and Kohonen Classification and Intervention Analysis. The models are trained and assessed on load data extracted from EUNITE network competition.

Text retrieval engines, such as search engines, always return a list of documents in response to a given query. Existing evaluations of text retrieval algorithms mostly use Precision and Recall of the returned list of documents as main quality measures of a search engine. In this paper, we propose a novel approach for comparing different algorithms adopted by different search engines and evaluate their performance. In our approach, the results of each algorithm is treated as an inter-related set of documents and the effectiveness of the algorithm is evaluated based on the degree of relation in the set of documents. After verifying the correctness of the evaluation measure by examining the results of the two retrieval algorithms, BM25 and pivoted normalization, and comparing these results with an ideal ranking, we compare the results of these algorithms and investigate the impact of certain major factors like stemming on the results of the suggested algorithm. The effectiveness of our proposed method is justified through obtained experimental results.

Polynomials are one of the most powerful functions that have been used in many fields of mathematics such as curve fitting and regression. Low order polynomials are desired for their smoothness1, good local approximation and interpolation. Being smooth, they can be used to locally approximate almost any derivable function. This means that when linear functions fail in approximation (e.g. where the first order Taylor expansion equals zero) polynomial functions can be used in local approximation, such that one can achieve better estimations at extremums. In this paper, application of polynomial kernel functions in locally linear neurofuzzy models is shown. Using polynomial kernels in local models, better local approximations in prediction of chaotic time series such as Mackey-Glass is achieved, and the capability of the neurofuzzy network is enhanced.

Classification is a well known task in data mining and machine learning that aims to predict the class of items as accurately as possible. A well planned data classification system makes essential data easy to find. An object is classified into one of the categories called classes according to the features that well separated the classes. Actually, classification maps an object to its classification label. Many researches used different learning algorithms to classify data; neural networks, decision trees, etc. In this paper, a new classification approach based on cooperative game is proposed. Cooperative game is a branch of game theory consists of a set of players and a characteristic function which specifies the value created by different subsets of the players in the game. In order to find classes in classification process, objects can be imagine as the players in a game and according to the values which obtained by these players, classes will be separated. This approach can be used to classify a population according to their contributions. In the other words, it applies equally to different types of data. Through out this paper, a special case in medical diagnosis was studied. 304 samples taken from human leukemia tissue consists of 17 attributes which determine different CD markers related to leukemia were analyzed. These samples collected from different types of leukemia at Iran Blood Transfusion Organization (IBTO). Obtained results demonstrate that cooperative game is very promising to use directly for classification.

In this paper we propose a new integration method for multi-feature object tracking in a particle filter framework. We divide particles into separate clusters. All particles within a cluster measure a specific feature. The number of particles within a cluster is in proportion to the reliability of associated feature. We do a compensation stage which neutralizes the effect of particles weights mean within a cluster. Compensation stage balances the concentration of particles around local maximal. So, particles are distributed more effectively in the scene. Proposed method provides both effective hypothesis generation and effective evaluation of hypothesis. Experimental results over a set of real-world sequences demonstrate better performance of our method compared to the common methods of feature integration.

Long-term forecasting of load demand is necessary for the correct operation of electric utilities. There is an on-going attention toward putting new approaches to the task. Recently, Neurofuzzy modeling has played a successful role in various applications over nonlinear time series prediction. This paper presents a neurofuzzy model for long-term load forecasting. This model is identified through Locally Linear Model Tree (LoLiMoT) learning algorithm. The model is compared to a multilayer perceptron and hierarchical hybrid neural model (HHNM). The models are trained and assessed on load data extracted from a North- American electric utility.

In this paper we propose a new approach for text region extraction in camera-captured document images. Texture-Edge Descriptor, TED, is utilized for text region extraction. TED is an 8-bit binary number which its bits are structural. This structural bits and special text region characteristics in document images make TED an appropriate descriptor for text region extraction. Applying well-known water flow method to the text regions extracted by TED, results in fast and good quality document image binarization. Experimental results demonstrate the effectiveness of our method for text region extraction and document image binarization.

In this paper, we present a novel hybrid algorithm for binarization of badly illuminated document images. This algorithm locally enhances the document image and makes the gray levels of text and background pixels separable. Afterward a simple global binarization algorithm binarizes the enhanced image. The enhancement process is a novel method that uses a separate transformation function to map the gray level of each pixel into a new domain. For each pixel, the transformation function is determined using its neighboring pixels gray level. The proposed binarization algorithm is robust for wide variety of degraded document images. Evaluation over a set of degraded document images illustrates the effectiveness of our proposed binarization algorithm.

In this paper, we present a novel hybrid algorithm for binarization of badly illuminated document images. This algorithm locally enhances the document image and makes the gray levels of text and background pixels separable. Afterward a simple global binarization algorithm binarizes the enhanced image. The enhancement process is a novel method that uses a separate transformation function to map the gray level of each pixel into a new domain. For each pixel, the transformation function is determined using its neighboring pixels gray level. The proposed binarization algorithm is robust for wide variety of degraded document images. Evaluation over a set of degraded document images illustrates the effectiveness of our proposed binarization algorithm.

In this paper, we present a novel hybrid algorithm for binarization of badly illuminated document images. This algorithm locally enhances the document image and makes the gray levels of text and background pixels separable. Afterward a simple global binarization algorithm binarizes the enhanced image. The enhancement process is a novel method that uses a separate transformation function to map the gray level of each pixel into a new domain. For each pixel, the transformation function is determined using its neighboring pixels gray level. The proposed binarization algorithm is robust for wide variety of degraded document images. Evaluation over a set of degraded document images illustrates the effectiveness of our proposed binarization algorithm.

In this paper, a fuzzy inference system by which reliability of features can be measured is designed. The reliability determines discriminative power of a feature in separating target from background. We focus our attention on design of membership functions. With a rational explanation on available information over a particle filter-base tracking process, we infer a coarse estimation of membership functions. It follows with a fine-tuning stage by using genetic algorithm. Color, edge, texture and TED are used in current work but the extension to a wider number of features is straightforward.

Background modeling is one of the most important parts of visual surveillance systems. Most background models are pixel-based which extract detailed shape of moving objects, but they are so sensitive to nonstationary scenes. In many applications there is no need to detect the detailed shape of moving objects. So some researchers use block-based methods instead of pixel-based which are more insensitive to local movements. These two methods are complementary to each other. We propose an efficient hierarchical method by which the block level information is utilized intelligently to improve the efficiency and robustness of pixel level. Experimental results demonstrate the effectiveness of the algorithm when applied in different outdoor and indoor environments.

Being able to predict the coming seizure can impressively improve the quality of the patients' lives since they can be warned to avoid doing risky activities via a prediction system. Here, a locally linear neuro fuzzy model is used to predict the EEG time series. Subsequently, this model is utilized in accompany with Singular Spectrum Analysis for prediction. Afterward, an information theoretic criterion is used to select a reliable subset of input variables which contain more information about the target signal. Comparison of three mentioned methods on one hand shows that SSA enables our prediction model to extract the main patterns of the EEG signal and highly improves the prediction accuracy. On the other hand, applying the method of channel selection to the model yields more accurate prediction. It is shown that fusion of some certain signals provides more information about the target and considerably improves the prediction ability.

Video processing techniques based on pattern recognition methods and machine vision is one of the interesting research fields which attract many researchers. In this paper, we proposed a novel method for video summarization using genetic algorithm based on information theory. Our method relies on the mutual information for video summarization. The information theory measure provides us with better results because it extracts the inter-frame information. We present that it is a suitable factor for summarizing video, which maintains its integrity.

In this paper, we address the Heart Sound signal modeling problem. The approach taken is based on sparse and redundant representations on an overcomplete dictionary. We apply matching pursuit (MP) and orthogonal matching pursuit (OMP) on two sets of normal and pathological phonocardiograms (PCGs). The dictionary includes classical Gabor wavelets or time-frequency atoms which are the product of a sinusoid and a Gaussian window function. The normalized root-mean-square error (NRMSE) was computed between the original and the reconstructed signals. The results show that the OMP method is very suitable to the transient and complex properties of the PCG’s, as it yielded excellent NRMSE’s around 1.61% for normal sounds and 5.19% for pathological murmurs.

Quantum Evolutionary Algorithm (QEA) is a novel optimization algorithm which uses a probabilistic representation for solution and is highly suitable for combinatorial problems like Knapsack problem. Fractal image compression is a well-known problem which is in the class of NP-Hard problems. Genetic algorithms are widely used for fractal image compression problems, but QEA is not used for this kind of problems yet. This paper uses a novel Functional Sized population Quantum Evolutionary Algorithm for fractal image compression. Experimental results show that the proposed algorithm has a better performance than GA and conventional fractal image compression algorithms.