Block Diagonal Representation (BDR) model can efficiently cluster data by using linear representation, but it cannot make good use of non-linear manifold information commonly appeared in high-dimensional data. To solve this problem, the improved Block Diagonal Representation based on Neighbor Graph (BDRNG) clustering algorithm was proposed to perform the linear fitting of the local geometric structure by the neighbor graph and generate the block-diagonal structure by using the block-diagonal regularization. In BDRNG algorithm, both global information and local data structure were learned at the same time to achieve a better clustering performance. Due to the fact that the model contains the neighbor graph and non-convex block-diagonal representation norm, the alternative minimization was adopted by BDRNG to optimize the solving algorithm. Experimental results show that:on the noise dataset, BDRNG can generate the stable coefficient matrix with block-diagonal form, which proves that BDRNG is robust to the noise data; on the standard datasets, BDRNG has better clustering performance than BDR, especially on the facial dataset, BDRNG has the clustering accuracy 8% higher than BDR.

Pedestrian Re-IDentification (Re-ID) is a hot issue in the field of computer vision and mainly focuses on “how to relate to specific person captured by different cameras in different physical locations”. Traditional methods of Re-ID were mainly based on the extraction of low-level features, such as local descriptors, color histograms and human poses. In recent years, in view of the problems in traditional methods such as pedestrian occlusion and posture disalignment, pedestrian Re-ID methods based on deep learning such as region, attention mechanism, posture and Generative Adversarial Network (GAN) were proposed and the experimental results became significantly better than before. Therefore, the researches of deep learning in pedestrian Re-ID were summarized and classified, and different from the previous reviews, the pedestrian Re-ID methods were divided into four categories to discuss in this review. Firstly, the pedestrian Re-ID methods based on deep learning were summarized by following four methods region, attention, posture, and GAN. Then the performances of mAP (mean Average Precision) and Rank-1 indicators of these methods on the mainstream datasets were analyzed. The results show that the deep learning-based methods can reduce the model overfitting by enhancing the connection between local features and narrowing domain gaps. Finally, the development direction of pedestrian Re-ID method research was forecasted.

To detect surface object information quickly and accurately by using remote sensing images is a current research hot spot. In order to solve the problems of inefficiency of the traditional manual visual interpretation segmentation method as well as the low accuracy and a lot of background noise of the existing remote sensing image segmentation based on deep learning in complex scenes, an image segmentation algorithm based on improved U-net network architecture and fully connected conditional random field was proposed. Firstly, a new network model was constructed by integrating VGG16 and U-net to effectively extract the features of remote sensing images with highly complex background. Then, by selecting the appropriate activation function and convolution method, the image segmentation accuracy was improved while the model prediction time was significantly reduced. Finally, on the basis of guaranteeing the segmentation accuracy, the segmentation result was further improved by using fully connected conditional random field. The simulation test on the standard dataset Potsdam provided by ISPRS showed that the accuracy, recall and the Mean Intersection over Union (MIoU) of the proposed algorithm were increased by 15.06 percentage points, 29.11 percentage points and 0.3662 respectively, and the Mean Absolute Error (MAE) of the algorithm was reduced by 0.02892 compared with those of U-net. Experimental results verify that the proposed algorithm is an effective and robust algorithm for extracting surface objects from remote sensing images.

Focusing on the issue of target motion feature extraction of video sequences in complex scene, and referring to the motion perception of biological vision system to the moving video targets, the traditional primary Visual cortex (V1) cell model of visual cortex was improved and a novel method of random-dot motion feature extraction based on the mechanism of biological visual cortex was proposed. Firstly, the spatial-temporal filter and half-squaring operation combined with normalization were adopted to simulate the linearity and nonlinearity of neuron's receptive field. Then, a universal V1 cell model was obtained by adding a directional selectivity adjustable parameter to the output weight, which solved the problem of the single direction selectivity and the disability to respond correctly to multi-direction motion in the traditional model. The simulation results show that the analog outputs of proposed model are almost consistent with the experimental data of biology, which indicates that the proposed model can simulate the V1 neurons of different direction selectivity and extract motion features well from random-dot video sequences with complex motion morphs. The proposed method can provide new idea for processing feature information of optical flow, extract motion feature of video sequence and track its object effectively.

The deformation such as translation, rotation and random scaling of local images in image recognition tasks is a complicated problem. An algorithm based on pre-training convolutional filters and Multi-Scale block Convolutional Neural Network (MS-CNN) was proposed to solve these problems. Firstly, the training dataset without labels was used to train a sparse autoencoder and get a collection of convolutional filters with characteristics in accord with the dataset and good initial values. To enhance the robustness and reduce the impact of the pooling layer for the feature extraction, a new Convolutional Neural Network (CNN) structure with multiple channels was proposed. The multi-scale block operation was applied to input image to form several channels, and each channel was convolved with corresponding size of filter. Then the convolutional layer, a local contrast normalization layer and a pooling layer were set to obtain invariability. The feature maps were put in the full connected layer and final features were exported for target recognition. The recognition rates of STL-10 database and remote sensing airplane images were both improved compared to traditional CNN. The experimental results show that the proposed method has robust performance when dealing with deformations such as translation, rotation and scaling.

Focusing on the problem of nonlinear time series prediction in the field of analog circuit fault prediction and the problem of error accumulation in traditional Support Vector Regression (SVR) multi-step prediction, a new adaptive SVR prediction algorithm based on phase space reconstruction was proposed. Firstly, the significance of SVR multi-step prediction method for time series trend prediction and the error accumulation problem caused by multi-step prediction were analyzed. Secondly, phase space reconstruction technique was introduced into SVR prediction, the phase space of the time series of the analog circuit state was reconstructed, and then the SVR prediction was carried out. Thirdly, on the basis of the two SVR prediction of the error accumulated sequence generated in the multi-step prediction process, the adaptive correction of the initial prediction error was realized. Finally, the proposed algorithm was simulated and verified. The simulation verification results and experimental results of the health degree prediction of the analog circuit show that the proposed algorithm can effectively reduce the error accumulation caused by multi-step prediction, and significantly improve the accuracy of regression estimation, and better predict the change trend of analog circuit state.

A Combined Prediction Scheme (CPS) and a concept of Prediction Accuracy Assurance (PAA) were put forward for the runtime of local and remote tasks, on the issue of inapplicability of the singleness policy to all the heterogeneous tasks. The toolkit of GridSim was used to implement the CPS, and PAA was a quantitative evaluation standard of the prediction runtime provided by a specific strategy. The simulation experiments showed that, compared with the local task prediction strategy such as Last and Sliding Median (SM), the average relative residual error of CPS respectively reduced by 1.58% and 1.62%; and compared with the remote task prediction strategy such as Running Mean (RM) and Exponential Smoothing (ES), the average relative residual error of CPS respectively reduced by 1.02% and 2.9%. The results indicate that PAA can select the near-optimal value from the results of comprehensive prediction strategy, and CPS enhances the PAA of the runtime of local and remote tasks in the computing environments.

The unexpected power loss or system crash can result in data inconsistency upon updating a persistent data structure. Most existing file systems use some consistency techniques such as write-ahead logging, copy-on-write to avoid this situation. These mechanisms, however, introduce a significant overhead, and fail to adapt to the diversity of directory and heterogeneity of data reliability demands. Existing file-adaptive journaling technique is required to modify legacy applications. Therefore, a directory-adaptive journaling mode selective mechanism for Android systems was proposed to choose different journaling modes with strong or weak consistency guarantees in terms of different directories reliability demands. This mechanism is transparent to developers, and also matches the feature of Android systems, hence, it greatly reduces the consistency guarantee overhead without sacrifice of reliability. The experimental results show that modified file system can identify directories in which a file resides, meanwhile, choose reasonable pre-defined journaling mode.

Aiming at the problem that the current display module could not provide a perfect stereo vision on the principle of human visual system, a solution of Virtual Reality (VR) stereo vision was proposed based on the oblique crossing frustum camera. Firstly, by studying the ken model and the theory of accessing to the depth data by eyes, a mathematical model of eyes parallex was built. Secondly, the industrial engine 3DVIA Studio was used as the simulation platform, which relied on the VSL programming language to screen. The relationship of child and parent was set up and the module of visual interaction was designed to construct the stereo camera. Then, the point cloud model was developed to quantize the stereo sense. The advantages and disadvantages of each model were analyzed based on the characteristics of depth display and distortion, and all models were optimized step by step. Center axis parallel normal frustum camera model and normal frustum model whose center axis crossed at the viewing distance were developed, the frustum of camera was optimized to develop a VR camera model of oblique crossing frustum. At last, using 3DVIA Studio as the experiment platform, specific data were substituted on it to do projective transformation. The result shows that the proposed camera model of oblique crossing frustum eliminates the distortion guarantees the depth information display effection, and provides an excellent effect of vision.

Concerning the Bit Error Rate (BER) and channel capacity optimization in the data transfer process of Multi-Input Multi-Output (MIMO) system, a multi-user MIMO multi-hop relay system based on cooperative likelihood detection and sphere decoding was proposed. Firstly, it constructed a second-order cooperative MIMO relay system model to analyze the relay transmission process of channel data, as well as path loss and shadow fading. Secondly, the sphere decoding was used to derive the equivalent maximum likelihood rule for log-normal shadow fading detection. Finally, the maximum channel power harmonic mean selection policy was put forward, and an access link with smaller BER was chosen for user based on correlation link metric and maximum channel power threshold value, so as to improve the performance of multi-user MIMO system. Simulation results show that, compared with the multi-user MIMO multi-hop relay system based on mutual information maximization and the relay system based on decoding-forwarding and MIMO orthogonal Space Time Block Code (STBC), the average BER of the proposed system reduced by 27.4% and 32.6% respectively, and the average channel capacity increased by 9.5% and 12.7% respectively. The results demonstrate that the proposed system has a good effect in reducing BER and improving channel capacity.

This paper proposed a novel sparse tracking method based on multi-feature fusion to compensate for incomplete description of single feature. Firstly, to fuse various features, multiple feature descriptors of dictionary templates and particle candidates were encoded as the form of kernel matrices. Secondly, every candidate particle was sparsely represented as a linear combination of all atoms of dictionary. Then the sparse representation model was efficiently solved using a Kernelizable Accelerated Proximal Gradient (KAPG) method. Lastly, in the framework of particle filter, the weights of particles were determined by sparse coefficient reconstruction errors to realize tracking. In the tracking step, a template update strategy which employed incremental subspace learning was introduced. The experimental results show that, compared with the related state-of-the-art methods, this algorithm improves the tracking accuracy under all kinds of factors such as occlusions, illumination changes, pose changes, background clutter and viewpoint variation.

Aiming at improving the robustness in pre-processing and extracting features sufficiently for Synthetic Aperture Radar (SAR) images, an automatic target recognition algorithm for SAR images based on Deep Belief Network (DBN) was proposed. Firstly, a non-local means image despeckling algorithm was proposed based on Dual-Tree Complex Wavelet Transformation (DT-CWT); then combined with the estimation of the object azimuth, a robust process on original data was achieved; finally a multi-layer DBN was applied to extract the deeply abstract visual information as features to complete target recognition. The experiments were conducted on three Moving and Stationary Target Acquisition and Recognition (MSTAR) databases. The results show that the algorithm performs efficiently with high accuracy and robustness.

Several limitations exist in the current spam filtering methods, such as they usually rely on only one type of E-mail characteristic to realize the E-mail classification, and have poor adaptability to the dynamic changes of E-mail characteristics. Concerning these limitations, a hybrid spam filtering method based on users' feedback was proposed. Based on the Social Network (SN) relationship among users, the dynamic update of the knowledge for E-mail classification was achieved with the help of the user's feedback scheme. Furthermore, the Bayesian model was introduced to integrate the content-based and the identity-based characteristics of E-mail in the classification. The simulation results show that the proposed method outperforms the traditional method in terms of E-mail classification, when the E-mail characteristics change dynamically. The overall recall, precision and accuracy ratios of the method can achieve 90% and above. While guaranteeing the performance of E-mail classification, the proposed method can improve the adaptability of classification to the changes of E-mail characteristics effectively. Therefore, the proposed method can act as a useful complement to the current spam filtering methods.

Traditional image object detection algorithm based on local feature is sensitive to rotation and occlusion; meanwhile, it also obtains low detection precision and speed in many cases. In order to improve the performance of this algorithm, a new image objects detection method applying objects’ local feature to sparse representation theory was introduced. Employing supervised random tree method to learn local features of sample images, a dictionary could be formed. The combination of sub-image blocks of test image and well trained dictionary in first stage could predict the location of the object in the test image, in this way it could obtain a sparse representation of the test image as well as the object detection goal. The experimental results demonstrate that the proposed method achieves robust detection results in rotation, occlusion condition and intricate background. What’s more, the method obtains higher detection precision and speed.

The particle system can hardly simulate the natural features effectively and real-time at the same time. Based on the analysis of traditional simulation methods, a real-time rendering algorithm was presented. The algorithm used rectangular elementary to simulate snow particles，added texture mapping based on superposition，normalized snow lifecycle for color blending，and regulated the size of snow particle and snow density according to the size of temperature. In the snow-falling stage, this algorithm introduced Level Of Detail (LOD) technology, imitated the movement effect according to the whereabouts of the snow particles moving force characteristics, took the force comprehensive consideration and simplified to reduce the computational complexity and improve the snow simulation authenticity.In the snow-accumulating stage, this method obtained the exposed surface of the scene model and its height field. And then based on a point coordinates and the point height in the field in a plane, it got the position of snow particles steadily to simulate the height change effect. This new method used the whole surface as ground-object emitter and superimposed the corresponding particle texture to improve authenticity.

As the fabric defect type of diversity and traditional artificial detection methods inefficient ,in order to detect the fabric defect more effective, A new approach, SGE, based on saliency region feature for fabric defect detection is studied. In this approach, the original image is divided into two parts, one extracts the saliency region feature of fabric defect by improved FSR roughly, another employing the gabor filter and taking the amplitude as an output characteristics, and extracts the saliency region feature of fabric defect by PSR accurately, then by using maximum entropy to segment the saliency region respectively and fused the sub-images. The result is get got by calculating perimeter and area of the contours to removal the isolated points. The experiment selects four types of typical fabric defect images and OpenCV library is used. The experiment result shows that the algorithm, without prior learning，meet the real-time.

Lots of spectrum data can not be identified or identified with low accuracy, especially in the case of large scale database, the former algorithm loses accuracy fastly. This paper presented a new blind search algorithm. This algorithm is based on a kind of brand-new score model based on similarity relationship measurement.For large scale question, the agorithm takes two pre-filtering strategies such as parention mass filtering and Peptide Sequence Tags (PST) filtering, so that it can guarantee the accurancy in large scale question.

Depth map is the key technology of “2D video+depth map” for 3D display. On the basis of the research into quaternion and Gabor filter, the depth map acquisition technique based on Quaternion-Gabor wavelet motion estimation was proposed. Through calculating the global motion vector of image from ordinary video, the background motion model was estimated and the motion field was gotten. In the end, the foreground and the background of the image were isolated, and the depth map of the image was obtained. Through expanding ordinary Gabor filter to Quaternion-Gabor filter, it can not only get extra information through transforming the picture to frequency domain, but also can get independent filter to each pixel's RGB. The experimental results show that the changes of depth map obtained by Quaternion-Gabor wavelet motion estimation will be very smooth and edges will be more outstanding.

In this paper, an adaptive color image enhancement method was proposed: Firstly, color image was transformed from RGB to HSV color space and the H component was kept invariable, while the illumination component of brightness image could be estimated through Adaptive Bidimensional Empirical Mode Decomposition (ABEMD); Secondly, reflection component was figured out by the method of center/surround Retinex algorithm, and the illumination and reflection components were controlled through Gamma emendation and Weber's law and processed with weighted average method; Thirdly, the S component was adjusted adaptively based on characteristics of the whole image, and then image was transformed back to RGB color space. The method could be evaluated by subjective effects and objective image quality assessment, and the experiment results show that the proposed algorithm is better in mean value, square variation, entropy and resolution than MSR algorithm and Meylan's algorithm.

Vector Quantization(VQ) is one of the popular codebook design methods for text-independent speaker identification. The key problem of VQ is the design of codebook. Speech feature parameters have complex distribution with high dimensions. Therefore, we have great difficulty in designing codebook. The traditional LBG algorithm yields only local optimal codebook. In this paper, a new method of codebook design was proposed, named as hybrid immune algorithm. It utilized the niche technology and K-means algorithm in the immune algorithm train step. It adopted improved mutation operator for data clustering with high dimension, reduced the blindness of stochastic mutation, so as to improve the local and global searching capability and increase the convergent speed by vaccination. Experiment for text-independent speaker identification shows that this method can obtain more optimum VQ parameters and better results than the LBG and hybrid genetic algorithm.

A semiautomatic method for medical image segmentation based on active shape model was introduced. In order to improve the segmentation speed and precision, a semiautomatic method was used to model the training set, and a Gaussian Pyramid of images with different resolutions was generated so that multi-resolution image search could be performed. Experiments and analysis show that this method can be used to segment the lung region of medical images and the results are quite better.

An algorithm of the extracting Lexicalized Tree Adjoining Grammar(LTAG) from Penn Chinese corpus was presented. Idea of the algorithm is to induce three kinds of trees from lexicalized tree bank. Then the method of Head-driven Phrase Structure Grammar(HPSG) was applied to extract lexicalized tree from corpus. In the end,invalid lexicalized trees were filtered out by linguistic rules. It requires fewer human efforts compared with hand-crafted grammar. It is possible to remedy omission of grammatical syntactic structures in hand-crafted grammar.