فصلنامه پردازش علائم و داده ها
سال سیزدهم شماره 3 (پیاپی 29، پاییز 1395)

In this paper, a new feature reduction technique for hyperspectral data classification problem is proposed based on extracting new features with smaller dimension with respect to the original data. For each pixel of a hyperspectral image, a specific rational function approximation is developed to fit its own spectral response curve (SRC) and the coefficients of the numerator and denominator polynomials of this function are considered as new extracted features. The method focuses on geometrical nature of SRCs and relies on the fact that the sequence discipline - ordinance of reflectance coefficients in spectral response curve - contains some information which has not been addressed by many other existing methods that are based on statistical analysis of data. Maximum likelihood classification results demonstrate that our method provides better classification accuracies compared to many competing feature extraction algorithms. In addition, the proposed algorithm has the possibility to be applied to all pixels of image individually and simultaneously.

The most important method of recognizing behavior of recurrent maps is to plot Bifurcation Diagram. Conventional method for plotting is to generate a variety of time series for different values of model parameter and plotting these points with respect to it after transient state. Bifurcation Diagrams are destitute of analytical interpretations for determining the cause of chaotic and period doubling behaviors. In this study, a diagram is presented which can be called Qualitative Bifurcation Diagram (QBD). One of the special features of Qualitative Bifurcation Diagram is recognition of parameter values and initial conditions, in which, the map has periodic, Aperiodic or chaotic behavior and for recognizing these behaviors, no primary information with regards to the map is needed.
The implementation of QBD on the logistic map indicates high susceptibility of the algorithm in distinguishing high periods and periodic windows; it has not only reduced the calculation time, but also has made it possible to examine initial conditions and logistic parameter effects with high velocity simultaneously. In spite of existence arrangement in Bifurcation Diagram, the result showed that logistic is not a mosaic arrangement, but it has a dynamic arrangement.
This article dealt with the investigation of Lyapunov Exponent as one of the tools of chaos examination which is analyzed analytically.

In this paper, a novel shape descriptor for shape based object retrieval is proposed. A growing process is introduced in which a contour is reconstructed from the bounding circle of the shape. In this growing process, circle points move toward the shape in normal direction until they arrive at the shape contour. Three different shape descriptors are extracted from this process: the first descriptor is defined as the number of steps that every circle point should pass which is called growing steps. The second descriptor is considered as the boundary distance of the circle points at the end of the growing process. The third descriptor is the mean of the curvature of the growing lines that are created by moving points. Invariance to translation is the intrinsic property of these features. By selecting a fixed starting point and tracing the boundary in a fixed direction (clock-wise or contour clock-wise), a set of descriptors can be collected that are invariant to rotation. Finally, normalizing the descriptors makes them to be invariant to scale. Support vector machines based on one-shot score are applied in the retrieval stage. Experimental results show that the suggested method has high performance for shape retrieval. It achieves 89.16% retrieval rate on MPEG-7 CE-Shape-1 dataset.

This paper presents a dynamic approach for skin detection- to separate the skin pixels from non-skin pixels- in color images. The static methods which use a fixed skin color model, fail if there is illumination variations or different skin colors in an image. Because of contextual information the proposed algorithm is less sensitive to the uncontrolled illumination conditions. In addition, the selection of discriminant features and the fusion of them and Bayesian classification increase the accuracy of the proposed method in comparison to the reference methods.

In this paper Intelligent image processing-based rapid method for detecting and tracking moving vehicles at intersections is proposed. In the detection part, the Gaussian mixture model used to obtain the moving parts. Then, the targets were detected using HOG features extracted from training images, Adaboost Cascade Classifier and the trained SVM.
At the tracking part, a number of key points on the image of the vehicle were identified at first. The center of mass of the object and the edges were used to obtain these key points because these points are primarily important and more common in tracking rigid bodies. Then, these points were tracked in consecutive frames using definitive adaptive procedures. Also, the Kalman filter was used to estimate new locations when the detector was not able to detect the targets.
The major advantage of this method compared to previous methods, is resistance against vehicle's overlapping and changes in Illuminations,so that the detection accuracy is 90.80% on overloaded traffic scenes and 88.75% accuracy on the tracking vehicles.

Question answering systems are produced and developed with the goal of providing short precise answers to questions asked in a natural language. Question classification is one the most important tasks in question answering systems. Question Classification is forecasting the type of response to given question in a natural language. Proposed methods divided into two categories: rule-based and machine learning-based. In this paper, a novel hybrid method for question classification was presented. The results of classifiers Combined By Voting, Behavior knowledge space, Naive Bayes, Decision Template and Dempster-Shafer Combination. The ensemble classifier exploits one rule-based classifier and two learning-based ones (SVM, Sparse Representation). The rule-based classifier includes a set of rules developed in the form of regular expressions. The learning-based classifications are based on lexical and syntactic features of questions. In the last part, the results of classification Combined By common methods in combination of one-class classifiers.

Credibility assessment screening by a small system and receiving optimum result in minimum time is a basic need in critical gates. Therefore the aim of this research is automatic detection of stress in guilty persons through skin conductance response and photoplethysmograph signals which are convenient and ease-of-use sensors .In this paper, a set of database with interview protocol (including control and relevant questions) in mock crime (Stealing jewels) is provided. 40 subjects participated in the experiments. 28 time-frequency features are extracted from two mentioned signals. The function of dimension reduction algorithms including principal component analysis, Kernel based PCA, linear discriminant analysis, cluster based LDA is evaluated to select optimum features. Support Vector Machine, Bayesian and AdaBoost are used as classifiers. The evaluation of algorithms on database is based on LOO method. Maximum accuracy (81.08%) is obtained through principal components analysis as feature selection method and Bayesian as classifier.

According to the researches, it turns out that human's activities are the results of the internal-neural activities of their brain. The reflection of such activities which are propagated throughout the scalp can then be acquired and processed. In this regard, brain signals can be acquired and recorded by EEG (Electroencephalography). Researchers have applied different technqiues for acquiring, pre-processing, feature extrcation and reduction and classifying EEG signal. According to published papers by Iranian researchers until 2015, it was found that most studies have been performed in medical applications and brain computer interface fields. Sampling and receiving EEG signals have been performed more in the central region than other regions. Statistical technqiues have more been used for feature extraction than other technqiues. Finally, the support vector machines are mostly used in the classification of brain signals. At the end, a study on anxiety and depression detection on fifty cases was performed in medical field. Simulation results show that our approach achieve an accuracy of up to 97 percent.

Spectral unmixing of hyperspectral images is one of the most important research fields in remote sensing. Recently, the direct use of spectral libraries in spectral unmixing is on increase by increasing the availability of the libraries. In this way the spectral unmixing problem is converted into a sparse regression problem that is time-consuming. This is due to the existence of irrelevant spectra in the library. So these spectra should be removed in some way. In the mentioned approach which is called sparse unmixing, we do not need an endmember extraction algorithm and determination of the number of endmembers priori. Since spectral libraries usually contain highly correlated spectra, the sparse unmixing approach leads to non-admissible solutions. On the other hand, most of the proposed solutions are not noise-resistant and do not reach to a sufficiently high sparse solution.
In this paper, with the purpose of overcoming the problems above, at first a machine learning approach for spectral library pruning is introduced. The spectral library is clustered using k-means algorithm based on a simple and efficient new feature space. Subtractive clustering is used for determining the cluster centers of k-means algorithm. Three distance measures, Spectral angle distance, spectral distance similarity and spectral correlation similarity tested to select the best for k-means. Then the training data needed to learn a classifier are extracted by adding different noise levels to the clustered spectra. The label of the training data is determined based on the results of spectral library clustering. After learning the classifier, each pixel of the image is classified using it. This classified image will be used for pruning the spectral library. For testing the impact of classifier type on the result of unmixing, three classifiers, decision tree classifier, neural networks and k-nearest neighbours are compared. For pruning the library, the spectra with the labels that none of the image pixels belong to, are removed from spectral library. In this way, the candidate spectra present in the image are extracted. Now, a genetic algorithm will be used for sparse unmixing.
Experimental results show that spectral library pruning enhances the performance of sparse unmixing algorithms. Moreover, using k-nearest neighbor in image classification step, gives better results in the final unmixing process. Genetic algorithm that used for sparse unmixing compared with OMP and SUnSAL algorithms, works well in noisy images.