Sharif Digital Repository

Segmentation of aerial images has been a challenging task in recent years. This paper introduces a simple and efficient method for segmentation and classification of aerial images based on a pixel-level classification. To this end, we use the Gabor texture features in HSV color space as our best experienced features for aerial images segmentation and classification. We test different classifiers including KNN, SVM and a classifier based on sparse representation. Comparison of our proposed method with a sample of segmentation pre-process based classification methods shows that our pixel-wise approach results in higher accuracy results with lower computation time  

During the last decade, there has been a growing interest toward the problem of sparse decomposition. A very important task in this field is dictionary learning, which is designing a suitable dictionary that can sparsely represent a group of training signals. In most dictionary learning algorithms, the cost function to determine the the optimum dictionary is the ℓ0 norm of the matrix of decomposition coefficients of the training signals. However, we believe that this cost function fails to fully express the goal of dictionary learning, because it only sparsifies the whole set of coefficients for all training signals, rather than the coefficients for each training signal individually. Thus,... 

The registration of images is a fundamental task in numerous applications in medical image processing. Similarity measure is an important key in intensity based image registration. Here, we propose correntropy measure as similarity measure in mono modal setting. Correntropy is a important measure between two random variables based on information theoretic learning and kernel methods. This measure is useful in non-Gaussian signal processing. In this paper, this measure is used in image registration. Here, we analytically illustrate that this measure is robust in presence of spiky noise (impulsive noise). The experimental results show that the proposed similarity has better performance than... 

In this paper, we present a new method to solve tracking problem. The proposed method combines sparse representation and motion estimation to track an object. Recently. sparse representation has gained much attention in signal processing and computer vision. Sparse representation can be used as a classifier but has high time complexity. Here, we utilize motion information in order to reduce this computation time by not calculating sparse codes for all the frames. Experimental results demonstrates that the achieved result are accurate enough and have much less computation time than using just a sparse classifier  

In this work we present a sparse dictionary learning method, specifically tuned to solve the tracking problem. Recently, sparse representation has drawn much attention because of its genuineness and strong mathematical background. In this paper we present an online method for dictionary learning which is desirable for problems such as tracking. Online learning methods are preferable because the whole data are not available at the current time. The presented method tries to use the advantages of the generative and discriminative models to achieve better performance. The experimental results show our method can overcome many tracking challenges  

Since tracking algorithms should be robust with respect to appearance changes, online algorithms has been investigated recently instead of offline ones which has shown an acceptable performance in controlled environments. The most challenging issue in online algorithms is updating of the model causing tracking failure because of introducing small errors in each update and disturbing the appearance model (drift). in this paper, we propose an online generative tracking algorithm in order to overcome the challenges such as occlusion, object shape changes, and illumination variations. In each frame, color distribution of target candidates is obtained and the candidate having the lowest distance... 

This paper presents a new transform domain to evaluate the goodness of fit of natural image data to the common Benford's Law. The evaluation is made by three statistical fitness criteria including Pearson's chi-square test statistic, normalized cross correlation and a distance measure based on symmetrized Kullback-Leibler divergence. It is shown that the serial combination of variance filtering and block 2-D discrete cosine transform reveals the best goodness of fit for the first significant digit. We also show that the proposed transform domain brings reasonable fit for the second, third and fourth significant digits. As an application, the proposed transform domain is utilized to detect... 

By increasing the number of registered speakers in Speaker Identification (SI) systems, computation time for identifying an unknown speaker is significantly increased. This problem arises from the simple design of conventional methods. Due to this limitation, we cannot use conventional SI methods in real time applications. In this paper, we propose a two-step method to overcome this limitation. We use different identification methods for each step. In the first step we reduce the search space using Nearest Neighbor method. In the second step we identify the target speaker using the conventional GMM-based SI method. The experimental results show 3.4× speed-ups without any accuracy loss using... 

Migraine is a common type of headache with neurovascular origin. In this paper, a quantitative analysis of spontaneous EEG patterns is used to examine the migraine patients with maximum and minimum pain levels. The analysis is based on alpha band phase synchronization algorithm. The efficiency of extracted features are examined through one-way ANOVA test. we reached the P-value of 0.0001, proving that the EEG patterns are statistically discriminant in maximum and minimum pain levels. We also used a Neural Network based approach in order to classify the EEG patterns, distinguishing between minimum and maximum pain levels. We achieved the total accuracy of 90.9 %  

This paper proposes four methods for improving the performance of keyword spotting (KWS) systems. Keyword models are usually created by concatenating the phoneme HMMs and garbage models consist of all phonemes HMMs. We present the results of investigations involving the use of skips in states of keyword HMMs and we focus on improving the hit ratio; then for false alarm reduction in KWS we model the words that are similar to keywords and we create HMMs for highly frequent words. These models help to improve the performance of the filler model. Two post-processing steps based on phoneme and word probabilities are used on the results of KWS to reduce the false alarms. We evaluate the... 

Scattered and little research in the field of Persian speech synthesis systems has been performed during the last ten years. Comprehensive framework that properly implements and adapts statistical speech synthesis methods for Persian has not been conducted yet. In this paper, recent statistical parametric speech synthesis methods including CLUSTERGEN, traditional HMM-based speech synthesis and its STRAIGHT version, are implemented and adapted for Persian language. CCR test is carried out to compare these methods with each other and with unit selection method. Listeners Score samples based on CMOS. The methods were ranked by averaging the CCR scores. The results show that STRAIGHT-based... 

Mel-frequency Cepstral coefficients are widely used for feature extraction in speech recognition systems. These features use Mel-scaled filters. A new filter-bank based on dependencies between frequency components and phoneme characteristics is proposed. F-ratio and mutual information are used for this purpose. A new filter-bank is designed in which frequency resolution of sub-band filters is inversely proportional to the computed dependency values. These new filterbank is used instead of Mel-scaled filters for feature extraction. A phoneme recognition experiment on FARSDAT Persian language database showed that features extracted using the proposed filter-bank reach higher accuracy (63.92%)... 

In noise radars, measuring the cross-correlation between transmitted and received signals is known as a common method of signal processing. However, it brings a lot of unwelcome sidelobes which can lean to masking weak echoes of far targets. There are many classical and modern methods of masking effect removal which are based on signal processing in the receiver side. A method of waveform design for decreasing this effect is presented and its efficiency is compared to purely random waveform generation method and CLEAN algorithm. Moreover, performance of collaborative using of receiver-based and transmitter-based techniques is considered  

Signal processing in noise radar is always done by calculating the cross-correlation between transmitted and received signals. Strong echoes of near targets present relatively high sidelobes in the correlation output, so they can conceal weak echoes of weak targets (masking effect). Many classical and modern methods for alleviating this effect have been offered, however, most of them are based on signal processing in the receiver side. In this paper, a new method of waveform design, able to decrease masking effect, is presented. Contrary to the other methods, this method keeps the matched filter structure in receiver unchanged and does not need performing any extra process in the receiver... 

This paper deals with blind separation of finite alphabet sources where we have n sources and only one observation. The method is applied directly in time (spatial) domain and no transformation is needed. It follows a two stage procedure. In the first stage the mixing coefficients are estimated, and in the second stage the sources are separated using the estimated mixing coefficients. We also study restrictions of this method and conditions for which its performance is acceptable. Simulation results are presented to show the ability of this method to source separation in images and pulse amplitude modulation (PAM) signals  

This paper uses singular value decomposition (SVD) for studying the dynamic properties of fatigue-loaded intervertebral disc. Previously, this problem had been addressed using mathematical models of using mass, spring and damper or based on poroelastic theory. This paper utilizes the signal processing approach and attempts to describe SVD based feature that can be an indicator for change in behavioral performance of the intervertebral disc warning the occurrence of temporary or permanent change in the structure or abnormality in behavior. The results are encouraging; however, further validation is required with more data  

In this study a new method is proposed for quantification of cardiac muscle motions in echocardiography frames based on empirical mode decomposition (EMD) and manifold learning method. EMD algorithm is able to extract intrinsic mode functions (IMF) from a signal. In the first bi-dimension intrinsic mode functions (BIMF) of echocardiography frames myocardial is shown with more details than the second BIMF and the second BIMF shows more details than the third BIMF. By using manifold learning method, quantification difference between each pair of consecutive frames in the first, second and third BIMF series (similarities between the frames were extracted). Acquired trajectories of three... 

A new approach to Image Quality Assessment (IQA) is presented. The idea is based on the fact that two images are similar if their structural relationship within their blocks is preserved. To this end, a transition matrix is defined which exploits structural transitions between corresponding blocks of two images. The matrix contains valuable information about differences of two images, which should be transformed to a quality index. Eigen-value analysis over the transition matrix leads to a new distance measure called Eigen-gap. According to simulation results, the Eigen-gap is not only highly correlated to subjective scores but also, its performance is as good as the SSIM, a trustworthy... 

Summary keywords are words that are used in the reference extracted summary, therefore can be used to discriminate between summary sentences from non-summary ones. Finding these words is important for the extractive summarization algorithms that measure the importance of a sentence based on the importance of its constituent words. This paper is focused on extracting summary keywords in the multi-party meeting domain. We test previously proposed keyword extraction algorithms and evaluate their performance to determine summary keywords. We also propose a new approach which uses discourse information to find local important keywords and show that it outperforms all the previous methods. We... 

We introduce the new concept of "metalines" for manipulating the amplitude and phase profile of an incident wave locally and independently. Thanks to the highly confined graphene plasmons, a transmit-array of graphene-based metalines is used to realize analog computing on an ultracompact, integrable, and planar platform. By employing the general concepts of spatial Fourier transformation, a well-designed structure of such meta-transmit-array, combined with graded index (GRIN) lenses, can perform two mathematical operations, i.e., differentiation and integration, with high efficiency. The presented configuration is about 60 times shorter than the recent structure proposed by Silva et al....