Xiaojun Jing

C-SURF: Colored Speeded Up Robust Features

SURF has been proven to be one of the state-of-the art feature detector and descriptor, and mainly treats colorful images as gray images. However, color provides valuable information in the object description and recognition tasks. This paper addresses this problem and adds the color information into the scale-and rotation-invariant interest point detector and descriptor, coined C-SURF (Colored Speeded Up Robust Features). The built C-SURF is more robust than the conventional SURF with respect to rotation variations. Moreover, we use 112 dimensions to describe not only the distribution of Harr-wavelet responses but also the color information within the interest point neighborhood. The evaluation results support the potential of the proposed approach.

Edge detection based on decision-level information fusion and its application in hybrid image filtering

A new edge detection method, based on decision-level information fusion, is proposed to classify image pixels into edge and non-edge categories. Traditional edge detection algorithms make the detection decision under a single criterion, which may perform inefficiently with a change of noise model. We use fusion entropy as a criterion to integrate decisions from different classifiers in order to improve the edge detection accuracy. The proposed decision fusion based edge detection method is applied to image filtering and leads to a weighted hybrid-filtering algorithm. Simulation results show that the new edge detection method has better performance than the single criterion edge detection methods.

Robust Collaborative Spectrum Sensing Using PHY-Layer Fingerprints in Mobile Cognitive Radio Networks

Collaborative spectrum sensing has been proposed to significantly improve the performance of spectrum sensing in cognitive radio networks (CRNs). However, a serious attack, called a primary user emulation attack (PUEA), could decrease the performance of collaborative spectrum sensing. In this letter, according to mobile CRNs, we propose a novel robust collaborative spectrum sensing method using physical-layer fingerprints power in a multipath Rayleigh fading channel. Moreover, a fingerprints-power-belief-based noncentral detection algorithm is designed to defend against PUEAs. Simulation results show that our proposed method can rapidly detect the presence of a primary user under PUEAs with good performance.

Palm vein recognition scheme based on an adaptive Gabor filter

We propose a novel palm vein recognition scheme based on an adaptive 2D Gabor filter. Three key steps were studied in this scheme: region of interest (ROI) extraction, adaptive Gabor filtering, and template matching. First, in the palm vein image extraction step, the authors used the index finger on both sides of the valley to locate the square area, and then iteratively expanded the area of the square box to maximise the ROI. Second, in the feature extraction step, a novel parameter selection scheme was proposed for optimising the Gabor filter. Third, in the template matching step, the author presented a novel template matching algorithm referred to as the minimum normalised Hamming distance. Experimental results demonstrated that the scheme achieved good performance with an EER of 0.12%.

Sentiment Analysis Using Modified LDA

The technology of the Internet develops rapidly recent years, the public tends to share their reviews, opinions and ideas on the Internet. The forms of these subjective texts are free and concise, and they contain a wealth of sentiment information. In this paper, a modified latent Dirichlet allocation (LDA) model and support vector machine (SVM) are used for sentiment analysis of subjective texts. Analysis of sentiment could help producer to enhance the products and guide user make better choices as well. We apply a modified LDA model using term frequency-inverse document frequency (TF-IDF) algorithm to mine potential topics, find the most relevant words of the topic and represent the document. Then we use SVM to categorize the texts into two classes: positive and negative. Experiment results show that the performance of the modified LDA approach is better than the traditional LDA model.

An Improved Blind Spectrum Sensing Algorithm Based on QR Decomposition and SVM

Spectrum sensing, a basic functionality in cognitive radio, aims at detecting the presence or absence of primary user (PU). As one of the most popular spectrum sensing methods, Covariance-based sensing works based on the correlation between signal samples. However, its performance sharply declines in low Signal Noise Ratio (SNR) environment. To improve detection performance of covariance-based sensing as far as possible, an improved blind spectrum sensing scheme is proposed in this paper on the basis of QR matrix decomposition and support vector machine (SVM). In the proposed scheme, QR matrix decomposition is applied to the co-variance matrix of received signal firstly, and then the main features are constituted by extracting and arranging orderly the upper triangular elements of R matrix. After that, SVM is used to conduct the obtained features and determine whether PU exists. The proposed algorithm does not need the prior information of PU and noise. Simulation results demonstrate that the proposed method has a better performance than conventional covariance-based methods, especially in low SNR scenarios.

A New Method of Spectrum Sensing in Cognitive Radio Based on Statistical Covariance Matrix

Spectrum sensing is a significant part of technique in a cognitive radio that detecting the presence of primary users in an authorized spectrum. the method that based on the statistical covariance matrix is one of main spectrum sensing techniques, using the difference of statistical covariance between the received signal and noise. In this paper, the new sensing method we proposed is also based on the statistical covariance. The new method compare to some traditional covariance algorithms has decrease the complexity of algorithm, at the same time, ensured the accuracy of detection. We give the statistics of detection, and we also find the threshold of the method when the probability of false alarm is given. The analysis and derivation process of threshold are provided in behind. Using Matlab for simulation to validate the correctness of the method and making the comparison with some typical detection method.

A Spectrum Sensing Scheme with Multiple Users

Spectrum sensing has attracted much concern of researchers due to its significant contribution to the spectral efficiency. However, the corresponding work mainly focuses on the sensing event of single primary users within a certain band and the investigation of the effect of PU traffic on the sensing performance is considered rarely. In this paper, a spectrum sensing scheme is presented to explore the co-existence of multiple users in the same frequency band based on subspace filtering. To remove uncertain noise as much as possible, subspace filtering is applied to the received signal of a cognitive radio, where the received signal is decomposed into two parts: noise subspace and signal-plus-noise subspace. Then the closed-form solution of the detection and false alarm probabilities with multiple users is given on the basis of the signal-plus-noise subspace in Rayleigh fading channel. Eventually, simulations are made to validate the proposed scheme.

Co-training Based on Multi-type Text Features

Sentiment classification is intended to classify the sentiment color categories expressed by the text. This paper illustrates the sentiment classification method based on the semi-supervised algorithm that aims to improve performance by using unlabeled data. This paper proposes a novel co-training style semi-supervised learning algorithm in order to improve semi-supervised learning ability. In our algorithm, there are three classifiers trained on the original labeled data, where the text representation for each classifier is unigram, bigram, and word2vec, respectively. And then these classifiers can use unlabeled data to update themselves. In detail, any of two classifiers have the same label, then add the new labeled data to a training set of the third classifier. By combining different types of features, our algorithm can extract text information from multiple views which contribute to sentiment classification. In addition, this algorithm doesn’t require redundant and sufficient perspectives. Experiments show that our algorithm is superior to traditional co-training algorithm and partial semi-supervised learning algorithm.

Performance Analysis for User-Centric Cloud Radio Access Network in Millimeter Wave

Millimeter wave (mmWave) and cloud radio access network (C-RAN) are two potential candidates for next generation communication. In this paper, we consider user-centric C-RAN in mmWave with the existence of blockages in urban areas. The remote radio heads (RRHs) are deployed according to a Poisson point process in the circular region ( {mathcal{D}} ), of radius R. We employ the stochastic geometry theory to analyze the signal-to-noise ratio (SNR), rate, and outage probability. We emphasize the effect of circular region radius on the performance in this network and evaluate the effect with Monte Carlo simulations. The simulation results show that SNR, rate and outage probability have the same asymptotic trends and have the best performance when replace the circular region ( {mathcal{D}} ) with the line-of-sight (LOS) circular region.