Search

Reducing the impact of hazy images on subsequent visual information processing is a challenging problem. In this paper, combining with atmospheric scattering model, we propose an end-to-end multi-scale feature multiple parallel fusion network called MMP-Net for single image haze removal. The MMP-Net includes three components: multi-scale CNN module, residual learning module and deep parallel fusion module. 1) In multi-scale CNN module, a multi-scale convolutional neural network (CNNs) is adopted to extract different scales features from whole to local, and these features are fused multiple times in parallel. 2) In residual learning module, residual blocks are introduced to deeply learn detailed features, which can recover more image details. 3) In deep parallel fusion module, those features from residual learning module are deeply merged with the fused features from CNNs, and finally used to recover a clean haze-free image via the atmospheric scattering model. The experimental results show that on the average of three datasets (SOTS, HSTS, and D-Hazy), proposed MMP-Net improves PSNR from 20.91db to 22.21db and SSIM from 0.8720 to 0.9023 over the best state-of-the-art DehazeNet method. What’s more, MMP-Net gains the best subjective visual quality on real-world hazy images.

Although deep learning-based hand gesture recognition techniques have achieved amazing performance in recent years, it is challenges as hand gesture is prone to interference, such as light changes and gesture shadows on gesture recognition. In this paper, we propose a recognition model based on ellipse skin color, and improved Deep Residual Learning network for gesture recognition. By utilizing the clustering effect of skin color in YCbCr color space, we segment CbCr two-dimensional space based on the elliptical skin color model to hand gesture. To further improve performance, the mathematical morphology with logic operation is proposed to assist hand gesture segmentation. Lastly, the Deep Residual Learning architecture is used to increase the width of the network and the channel attention mechanism. The experimental results show that the proposed method can obtain a high recognition rate compared with existing methods.

Aiming at the problem of mobile robot path planning in complex environment, a method of integrating fuzzy logic with artificial potential field (APF) was presented. To overcome the local minima problem of APF, a heading fuzzy controller was designed. While the robot is falling into the local minima, the fuzzy controller will generate an angle to modify the current heading so that the robot escapes local minima and avoid obstacles. According to the complexity of the working environment, a step fuzzy controller was designed, which enables the robot to adjust the step adaptively and reduces the algorithm loop times and computation time efficiently. Simulation results show that this method is effective and practical for the mobile robot path planning.

Stereoscopic image quality assessment (SIQA) is an essential and tricky part of image processing. Recently, many scholars have conducted image quality assessment in either spatial or frequency domain respectively. In this paper, we propose a no-reference stereoscopic image quality evaluation method by considering both complex contourlet and spatial features. Firstly, the original views are converted to the Lab color space. Then the luminance channel of Lab color space and the synthetic cyclopean image are calculated from the original stereo pairs. Next, the perceptual features of these types of images are extracted in the complex contour domain and the spatial domain respectively. Finally, all pre-extracted features are sent to the regression model for training and predicting image quality scores. Our experimental results show that the proposed algorithm achieves high consistency with human subjective perception, and compete with state-of-the-art algorithms.

A six-degree-of-freedom parallel manipulator system based on 6-PURU structure is developed for vehicles' motion simulation. The main research in this paper is focused on dynamic modeling of the proposed parallel system. The kinematic model of the proposed system is established via kinematic influence coefficient method, in which the velocity and acceleration mappings between the passive components and actuated components are more accurate compared to the equivalent 6-PUS or 6-PSS systems. Based on the kinematics analysis, the dynamic model of the parallel kinematic machine is then formulated via the principle of virtual work. Finally a numerical calculation is performed based on the proposed dynamic model which matches the simulation results of commercial software. The approach outlined in this paper is generic and can be extended to a variety of applications.

Three-dimensional (3D) reconstruction is of great significance to the research of spatial structure and physical properties of porous media. In this paper, we propose a kind of new 3D reconstruction algorithm of coal-rock microscopic image by combing pattern density function and genetic algorithm. One single 2D microscopic image is used as training image (TI) and pattern density function is used as constraint condition of genetic algorithm, and 3D structure of coal-rock is reconstructed by genetic algorithm. Our experimental results suggest that reconstructed results are similar with real structure at both physical and visual properties.

In recent years, convolutional neural networks have achieved more outstanding results and been widely used in the field of image quality assessment compared with the traditional handcraft method. This paper presents a no-reference image quality assessment method based on dual-channel convolutional neural network. The raw image is labeled by visual information fidelity and divided into multiple patches as input. After that, feature extraction is performed by two network channels with different pooling layers. The features are Iinearly stitched and sent to the fully connected layer. The experimental results on the LIVE database and the TID2008 database show that our model has the state-of-the-art performance and obtain a better consistency with human subjective assessment.

The equivalent sources reconstructed from near field has been used to predict the radiation characteristic in equivalent modeling of electromagnetic compatibility. In most existing source reconstruction methods, the ground plane of original radiator was assumed to be infinitely large. It means that the diffraction of ground plane edges is ignored, which may result in inaccuracy for equivalent modeling. In this paper, the effect of finite ground plane on the near electric field is studied first. Then a modified Green’s function which includes the edge diffraction effect is proposed to reconstruct the equivalent sources. A simple radiator consisting of magnetic dipoles above a ground plane is modeled to validate the proposed method.

In this paper, a tracking algorithm based on collaborative learning Kernelized Correlation Filter is proposed. According to the framework of KCF, two trackers based on HoG and CN are cooperatively trained, and independent scale adaptive algorithm is used at the same time, and we call it collaborative learning Kernelized Correlation Filter (CL-KCF) algorithm. Our experimental results show that the proposed algorithm has higher tracking accuracy and robustness than other trackers, and can automatically adapt to the target's scale change.

Most of the sea surface images are characterized by sky and sea water, and the color is monotonous so that the haze is often unevenly distributed, so the sea surface images dehazing are especially challenging. This paper presents a novel multi-scale cascaded network structure that can directly restore clear images from original hazy images without estimating the transmission map. The dehazed image is clearly sharper, so that the target on the sea can be better identified. The dehazing performances of our proposed method are evaluated on synthetic images as well as real sea surface images and also compared with other state-of-the-art techniques.

In this paper, we propose a general method based on variational inference for the system parameter (SP) identification of quantized finite impulse response (FIR) systems. The core is to first express the transition probability from actual data to quantized output with a simple general Gaussian process characterized by compensation parameters (CPs). Then the SP and CPs are iteratively estimated by proposing an expectation maximization (EM) based on variational inference (EM-VI). The EM-VI avoids the complex integral calculation, but with good SP identification performance. Finally, the simulation demonstrates that the EM-VI is feasible and effective with the almost same accuracy as the existing algorithms, but requires the less computation.

In this paper, we present a new nonlinear filter based on variational inference, which we have named the model fitting filter (MFF). The core of MFF is to represent or fit the nonlinear measurement likelihood probability with a linear Gaussian probability (LGP) characterized by some parameters that need to be turned. Further, MFF is analytically established on LGP, which is a little similar to the simple linear system filter. Specifically, the nonlinear filtering performance is well improved by utilizing these turned parameters to correct the state estimation. At the same time, MFF does not require the square root decomposition of covariance matrix, so it avoids the inborn defect of numerical instability, which exists more or less in UKF or DDF. The superior performance of MFF is numerically tested by a target tracking problem with nonlinear models.

The reconstruction of sparse radiation sources above a finite ground plane is important for the equivalent modeling of radiators such as slot antennas, printed circuit boards and slotted shields. Existing methods usually assume the infinitely large ground plane which is not realistically realizable and the results are not practically accurate since they ignore the edge diffraction effect of a realistic ground plane. In this paper, a finite ground plane is considered and its edge diffraction effect is modeled by the uniform theory of diffraction (UTD). A matrix equation relating sources with near fields is derived with the UTD and the half space Green's function approach. By solving the matrix equation, the radiation sources can be reconstructed from the near fields. In order to improve the efficiency and accuracy, the sparse source reconstruction is developed to solve the matrix equation. Different from existing methods, the proposed method considers the edge diffraction effect and applies the sparseness of the sources when solving the inverse problem. Therefore, it is more efficient and accurate than existing methods. Numerical results are presented to validate the effectiveness of the proposed method.

In this paper, we propose an AdaBoost Random Forest (AdaBoost-RF) based blind multiply distorted image quality assessment (IQA) method. The AdaBoost-RF is an ensemble learning algorithm that uses the principle of the AdaBoost and random forest as Weak Learners. We also operate learning quality-aware features (LQAF) and utilize the AdaBoost-RF to get the image quality score by taking advantage of image features. The AdaBoost-RF increases the degree of difference between Weak Learners, in contrast to other regressions, and can gain higher predictive accuracy. On the LIVE multiply distorted image database (LIVEMD) and MDID2013, experimental results show our proposed model has a better performance than the other mainstream IQA methods, and is a useful and reliable method for image quality assessment.

Multiview canonical correlation analysis (MCCA) is an effective tool for analyzing the relationships among group- aligned multidimensional samples, which has been applied to the fields of pattern recognition and computer vision. In MCCA, its first-stage canonical variables are solved by a multivariate eigenvalue problem that can be computed by Horst method. However, how to use the algorithm for effectively and stably solving higherstage projection directions is now still not clear. In this paper, we propose a multiview canonical variates learning algorithm, which uses a symmetric deflation strategy instead of asymmetric one for multi-set solutions. Also, we prove the convergence property of the proposed algorithm. Clearly this benefits the theoretical development of MCCA and can facilitate its applications in practice.

Most of the existing algorithm of objective image quality evaluation are often for a single type of distortion, and the effect of multi-distortion image quality evaluation is poor, In this paper presents a no reference image quality assessment method based on phase congruency and convolution neural network, to evaluate the mixed distorted image, Firstly, the input image is divided into blocks and phase congruency, and then the quality score of the image is trained and predicted by using the convolution network, The convolutional network structure consists of 4 layers of convolutions, 3 layers of maximum pooling, and 2 fully connected layers, The experimental results on the LIVE mulitipy distortion quality evaluation database show that the proposed method has a good consistency between the image quality and the subjective quality score

Using the data of A-share enterprises in the years between 2004 and 2012, this study investigates the relationship among the fluctuation of monetary policies, internal control, and financial constraints. The results show that although financial constraints are common among the firms, financial constraints tend to be strengthened by monetary contractions, and internal control could mitigate the positive effect of monetary contractions of financial constraints. After distinguishing between firms in different external environments and internal characteristics, we find that under monetary contractions, internal control could have greater mitigating effect on financial constraints in the enterprises which are located in areas with low degree of marketization and in private enterprises than in the enterprises which are located in areas with high degree of marketization and in state-owned enterprises. This study confers a new perspective on the determinants of financial constraints, provides evidence for the credit transmit channel of monetary policy, and gives empirical support to the internal control building of companies in our country.

Along with the development of split-share reform and capital market, the trend of equity decentralization in listed companies is becoming more and more obvious. Professional managers' betrayal of large shareholders frequently occurs, which leads to the competing for corporate control between executives and shareholders. This paper studies on the corporate control from the perspective of competing for corporate control between executives and shareholders. We analyze the executives' means of competing and try to find out how executives possess more corporate control by competing for board seats, uniting inside directors and transactional institutional investors.

We propose a new learning quality-aware features (LQAF) blind image quality assessment (IQA) algorithm for multiply distorted images. In the new model, some relevant quality perceptual features, including mean value of intensity, contrast sensitivity function (CSF), mean value of gradient magnitude, and 15 texture parameters of gray level-gradient co-occurrence matrix (GGCM) from four categories images: original distorted image and its phase congruency (PC) image, covariance maximum and minimum image of phase congruency, are used. Image quality estimation is accomplished by the approximating function between these features and subjective mean opinion scores using support vector regression (SVR). Experimental results on the LIVE multiply distorted image database (LIVEMD) demonstrate the effectiveness of our proposed method.

As discrete wavelet transform (DWT) is short in shift invariance and directionality, we proposed a no-reference image quality assessment (IQA) method based on dual-tree complex wavelet transform (DTCWT). In this method, an image is decomposed by DTCWT firstly. Then the energy value of each sub-band is calculated. Finally a support vector regressor (SVR) is adopted to predict image quality score. Experimental results show that our method is consistent well with human perception and has a low computational complexity.

The popularity of online content is unequally distributed. It has been observed that the top 20% most popular content account for about 80% of the network traffic. Inspired by the above observation, dynamic broadcast is designed to delivery popular content over broadcast networks to reduce the load of broadband network. Thus, identifying popular content is critically important. In this paper, we propose a new metric called Lifetime Popularity Score (LPS) to measure the popularity of online videos. The value of LPS is computed with a learned stochastic model, which establishes the relationship between video's historic view counts and its lifetime view counts. We validate the proposed approach on a real world data set of IPTV's VOD watch record and the experimental results have demonstrated the effectiveness of the proposed metric.

In this paper a new algorithm for recognizing digital numbers on coal gas meters is presented. Firstly, the RGB image of the coal gas meter is converted into the CIE 1931 XYZ color space, and the digital area is extracted in the CIE space by four line equations. Secondly the horizontal and vertical edges are detected on the G value image of the original RGB image, and tilt correction is finished by using the Radon transformation algorithm. Thirdly the CMYK color features are used to strengthen the contrast, and the digital regional characteristics are enhanced by using morphological operations. Fourthly, the ideal binary image is obtained by combing the `Otsu method' with a ratio estimation method, and then row and column projections are done to finish the segmentation of the digital numbers. Finally a neural network model trained using the back-propagation algorithm is adopted to recognize the numbers. Experimental results show that our method can attain ideal recognition results.

This paper presents a new BPNN model for image restoration by using area information around center pixel from degraded image and its edge image. This method gains edge image of blurred image by Sobel operator, and then uses BPNN model to build nonlinear restoration mapping relation by area pixel information from original and edge image. The edge information is extracted as a priori knowledge to recover the details and reduce the ringing artifact. Experimental results show that our proposed method could get more satisfying results than previous BPNN methods.

The task of Web session clustering is to group Web sessions according to similarities, so as to maximize the similarities within the group and, minimize the similarities between the groups. The number of clusters, the initial data points of the respective clusters, and the defining of criterion function are the 3 key points and difficulties that deserve consideration in Web session clustering. WSCBIS, Web session clustering based on increase of similarities, defines the number of clusters according to the knowledge of application fields; it takes advantage of ROCK to decide the initial data points of each cluster; it also determines the criterion function according to the contributions of overall increase in similarities made by dividing Web sessions into different clusters - which not only overcomes the shortcomings of traditional clustering algorithm which merely focus on partial similarities, but also decreases the complexities of time and space.

The task of clustering Web sessions is to group Web sessions based on similarity and consists of maximizing the intra-group similarity while minimizing the inter-group similarity. The first and foremost question needed to be considered in clustering Web sessions is how to measure the similarity between Web sessions. However, there are many shortcomings in traditional measurements. This paper analyses the shortcomings of traditional methods and introduces a new method to measure similarities between Web pages, which considers not only the URL but also the viewing time of the visited Web page. Then we propose a new method to measure the similarity of Web sessions using sequence alignment and the similarity of Web page access. Finally, we conclude this paper and propose the future research directions.