Trajectory data are crucial in intelligent transportation management, road network optimization, and urban mobility analysis. Many downstream applications, such as trajectory prediction and travel time estimation, rely on high-resolution trajectory data. However, real-world trajectories are often sparse due to GPS signal loss and power constraints. Existing trajectory recovery methods often struggle to utilize the latent hierarchical traffic conditions, and they often overlook complex movement semantics. To address these limitations, we propose sparse trajectory recovery with hierarchical dynamic traffic pattern inference (STREAM), a unified framework that collectively infers latent global and local traffic conditions from observed trajectories. By modeling these multi-scale dependencies in its encoder, STREAM enables the decoder to accurately reconstruct missing trajectory points. Additionally, our model effectively captures multi-step movement patterns to enhance the accuracy of next-location inference. Extensive experiments on real-world datasets demonstrate that our model outperforms nine existing competitors with an average improvement of 42.52% in trajectory recovery.

Millimeter-wave (mmW) automotive radar imaging technology shows significant promise in advanced driver assistance systems (ADAS). Super-resolution imaging methods can be employed the limited aperture length of automotive radar to improve azimuth (angular) resolution. However, automotive radar images typically exhibit large dynamic range (LDR) and large scene (LS), leading to pay extensive computational complexity and storage demands when striving for higher image quality. To tackle this challenge, a fast$l_{q}$sparsity-driven imaging method with adaptive-focusing framework (FLSD-AF) for mmWave automotive radar super-resolution imaging in this article. First, in AF framework, a detect-before-imaging (DBI) is proposed to make echo data to adaptive focused on potential target area (PTR), thereby reducing the dimension of the effective data to reduce computational complexity and storage demands. Second, a subspace-phase-compensation (SPC) is proposed to reduces storage demands of the measurement matrix by addressing the imaging model mismatch in near-field under LS. Finally, a fast$l_{q}$sparsity-driven (FLSD) imaging method is proposed. It employs$l_{q}$-norm nonconvex penalty function to address the biased problem to improve imaging quality under LDR, meanwhile the computational complexity of the matrix operation is greatly reduced by utilizing joint Kailath-Variant (K-V) formula and Gohberg-Semencul (G-S) factorization. In summary, the proposed FLSD-AF not only substantially enhances the imaging performance, but also significantly diminishes the storage demands and computational complexity under LDR and LS. The results of simulations and experimental data all verify the proposed method.

Ship detection in synthetic aperture radar (SAR) images is a hot pot. However, most existing convolution neural network (CNN)-based research is limited to single polarization ship detection and neglects the utilize of the rich polarization information to further improve detection performance. Thus, to address the problem, in this paper, a group-wise shuffle attention R-CNN (GWSA R-CNN) is proposed for ship detection in dual-polarization SAR images. Based on the raw Faster R-CNN, GWSA R-CNN embeds a group-wise shuffle attention module (GWSA module) in the detection subnetwork to capture enriched organic fusion polarization information. Finally, the experimental results on the dual-polarization SAR ship detection dataset (DSSDD) show the state-of-the-art (SOTA) performance of our GWSA R-CNN, outperforming than other 7 competitive models. Specifically, GWSA R-CNN surpasses the second-best model 1.82% average precision (AP).

We report a saliency-guided attention-based feature pyramid network (SA-FPN) for ship detection from synthetic aperture radar (SAR) images. The two key contributions are – 1) the saliency-guided technique and 2) the attention-based means. The former offers one unsupervised visual saliency map that can guide FPN to focus more on regions of interest (ROIs). The latter offers one supervised non-local feature self-attention map that can improve FPN’s global representation ability. We offer an effective combination scheme of the two. Experimental results on the open SSDD dataset reveal SA-FPN’s advanced SAR ship detection performance. Furthermore, the ablation studies can confirm the two contributions' effectiveness.

Ship detection from Synthetic Aperture Radar (SAR) images is of great importance. However, the diversity of ship target scales increases the difficulty of detection. To solve this problem, we propose a novel FPN which is enhanced by de-formable convo-lution, called Deform-FPN. Deformable convolution realizes multi-scale adaptive geometric deformation modeling of ships, and can extract multi-scale features of ships with strong ex-pression ability. The multi-level deformable convolution layers enhance the feature extraction and feature fusion capabilities. Specifically, we add deformable convolution to the backbone and lateral connection of Deform-FPN to improve the feature extraction ability. Experimental results on the SAR ship detec-tion dataset (SSDD) reveal the state-of-the-art performance of Deform-FPN, in contrast to other methods based on convolu-tional neural network (CNN). The experimental results show that Deform-FPN offers a 56.5% mAP that is superior to the suboptimal model DCN by 1.5%. In addition, we conducted ablation experiments to verify the effectiveness of the structure of the Deform-FPN we proposed.

Video synthetic aperture radar (Video SAR) has drawn much attention because it can continuously observe and track the moving target. Rather than tracking the target directly, it is better to track its shadow because the shadow has no location shift, and the back-scattering characteristic is stable. However, most current shadow tracking methods not only suffer from false alarms because their discrimination capacities are not good enough but also suffer from missed detection because the feature extraction capacities are limited under complicated environment. Therefore, we propose a shadow-enhanced self-attention and anchor-adaptive network (SE-SA-AAN) to achieve accurate moving target tracking for Video SAR. Firstly, the pre-processing technique sparse low-rank noise decomposition (SLRND) is proposed for enhancing shadows’ salience to facilitate subsequent feature extraction. Secondly, the transformer self-attention mechanism (TSAM) is embedded in the parameters-shared backbone in the feature extraction network to concentrate on regions of interests for suppressing clutter interference. Then, the representative features are input to the detector and tracker. The detector adds the semantic guided anchor-adaptive mechanism (SGAAM) to obtain optimized anchors that match the shadows’ location and shape in each frame. Meanwhile, the tracker applies a Siamese network to achieve trajectory tracking for each shadow. Based on the detection and tracking results, a data association is applied to achieve moving targets tracking. Finally, experiments on Sandia National Laboratories (SNL) data demonstrate that SE-SA-AAN outperforms the state-of-the-art methods FairMOT, TransTrack and Centertrack by 6.4%, 7.8% and 8.3% multiple object tracking accuracy (MOTA) separately.

Millimeter-wave (mmW) automotive radar imaging technology has shown significant potential in autopilot assistance systems. The automotive radar with limited aperture can achieve high-resolution image by synthetic aperture technology. However, conventional imaging methods result in strong background clutter and high sidelobe interferences. To solve these problems, we propose a target-oriented Bayesian compressive sensing imaging method with region-adaptive extractor (TO-BCS-RAE) for mmW automotive radar imaging. (1) First, to extract the potential-target-regions (PTR) as well as subtracting the background clutters outside the PTR in a high-resolution initial image (by synthetic aperture), a region-adaptive extractor (RAE) is developed with utilizing 2D CFAR, isolated-point removing, and imaging clustering. Meanwhile, a more accurate prior distribution of target scattering points can be obtained in the PTR. (2) Then, to suppress the background clutters while enhancing the smooth structure of targets in the PTR, a target-oriented Bayesian compressive sensing (TO-BCS) imaging method is proposed by combining the prior probability distributions and inherent continuity of the target scattering points. It can also effectively reduce the high sidelobes. (3) Finally, to verify the effectiveness of TO-BCS-RAE, we conduct experiments on real data collected from an automotive radar with a vehicle platform in three typical driving scenarios. Both simulated and experimental results show the imaging quality of the proposed imaging method over conventional BP, OMP and ISTA methods.

With the booming of Convolutional Neural Networks (CNNs), CNNs such as VGG-16 and ResNet-50 widely serve as backbone in SAR ship detection. However, CNN based backbone is hard to model long-range dependencies, and causes the lack of enough high-quality semantic infor-mation in feature maps of shallow layers, which leads to poor detection performance in complicated background and small-sized ships cases. To address these problems, we pro-pose a SAR ship detection method based on Swin Trans-former and Feature Enhancement Feature Pyramid Network (FEFPN). Swin Transformer serves as backbone to model long-range dependencies and generates hierarchical features maps. FEFPN is proposed to further improve the quality of feature maps by gradually enhencing the semantic infor-mation of feature maps at all levels, especially feature maps in shallow layers. Experiments conducted on SAR ship de-tection dataset (SSDD) reveal the advantage of our pro-posed methods.

Researchers can gain more intuitive information by converting synthetic aperture radar (SAR) images to optical images using generative adversarial networks (GANs). However, their GANs have poor feature extraction ability, which leads to color conversion errors and loss of details. Therefore, to solve this problem, we add an atrous spatial pyramid pooling (ASPP) module to GAN to enhance the feature extraction ability, i.e., ASPP-GAN. ASPP module can extract multi-resolution feature responses, enabling the network to focus on both overall and detailed features for better feature extraction ability. The experimental results on public SEN1-2 datasets show that ASPP-GAN has a significant improvement over the traditional GAN, i.e., Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) values are improved by about 20%.

Video synthetic aperture radar (SAR) has been found to be very valuable for detecting and tracking moving targets and observing areas of interest. Shadows produced by target motion in sequential radar images can be used to detect targets themselves. Since existing deep learning shadow detection methods often require many hand-designed components, in this paper, we propose a shadow detection method for video SAR moving target based on transformer, which is named Deformable Shadow-DETR. Deformable Shadow-DETR can better extract shadow features, and use the transformer encoder-decoder network to treat shadow detection as a direct set prediction problem, eliminating the need for cumbersome hand-designed components. Experiments on the real video SAR data published by the Sandia National Laboratories show that our proposed moving target shadow detection method can achieve excellent performance.

Recently, in the maritime monitoring field, ship detection in synthetic aperture radar (SAR) images has attracted increasing attention. Considering the characteristics of SAR images with small ship size and large ship aspect ratio, it is necessary for existing anchor boxes-based ship detection algorithm to generate anchor boxes matching the ground-truth boxes closer. Therefore, to tackle this problem, based on You Only Look Once version 5 (YOLOv5), we propose a K-means cluster method based on ship shape distance measure (SSD-Kmeans) for SAR ship detection. Aiming at anchor boxes clustering, SSD-Kmeans fully utilizes ship shape distance measure (i.e., length, width and aspect ratio) of SAR images to generate superior anchor boxes. In addition, SSD-Kmeans does not increase the model complexity of raw algorithm. Experimental results on Large-Scale SAR Ship Detection Dataset-v1.0 (LS-SSDD-v1.0) show that YOLOv5 with SSD-Kmeans can make 2.14% Average Precision (AP) improvement than YOLOv5 with K-means.

We propose an enhanced mask interaction network (EMIN) for ship instance segmentation from synthetic aperture radar (SAR) images. EMIN adopts three techniques to improve SAR ship instance segmentation performance — 1) an atrous spatial pyramid pooling (ASPP) to enable multi-resolution feature responses, 2) a non-local block (NLB) to capture long-range spatial dependencies, and 3) a concatenation shuffle attention (CSA) to boost mask interaction benefits. Results on the public SAR ship detection dataset (SSDD) show that — 1) the above each technique can offer an observable accuracy gain, and 2) EMIN surpasses the original MIN by 2.1% detection AP and 2.4% mask AP on SSDD.

Moving target shadows among video synthetic aperture radar (Video-SAR) images are always interfered by low scattering backgrounds and cluttered noises, causing poor detection-tracking accuracy. Thus, a shadow-background-noise 3D spatial decomposition (SBN-3D-SD) model is proposed to enhance shadows for higher detection-tracking accuracy. It leverages the sparse property of shadows, the low-rank property of backgrounds, and the Gaussian property of noises to perform 3D spatial three-decomposition. It separates shadows from backgrounds and noises by the alternating direction method of multipliers (ADMM). Results on the Sandia National Laboratories (SNL) data verify its effectiveness. It boosts the shadow saliency from the qualitative and quantitative evaluation. It boosts the shadow detection accuracy of Faster R-CNN, RetinaNet and YOLOv3. It also boosts the shadow tracking accuracy of TransTrack, FairMOT and ByteTrack.

This letter proposes a squeeze-and-excitation Laplacian pyramid network with dual-polarization feature fusion (SE-LPN-DPFF) for ship classification in synthetic aperture radar (SAR) images. SE-LPN-DPFF offers three contributions: 1) dual-polarization (VV and VH) feature fusion (DPFF); 2) channel modeling by the squeeze-and-excitation (SE) to balance each polarization feature’s contribution; and 3) Laplacian pyramid network (LPN) to achieve multiresolution analysis (MRA). Extensive ablation studies can confirm the effectiveness of each contribution. Results on the three- and six-category OpenSARShip datasets reveal the state-of-the-art SAR ship classification performance.

Existing deep learning (DL)-based synthetic aperture radar (SAR) ship instance segmentation models mostly extract feature subsets at the single level of feature pyramid network (FPN), and also ignore context information of the region of interest (ROI), which both hinder accuracy improvements. Thus, a full-level context squeeze-and-excitation ROI extractor (FL-CSE-ROIE) is proposed to handle these problems. FL-CSE-ROIE has three novelties: 1) full-level, i.e., extract feature subsets at each level of FPN to retain multi-scale features; 2) context, i.e., add multi-context surroundings of different scopes to ROIs to ease background interferences; and 3) squeeze-and-excitation (SE), i.e., balance contributions of different scope contexts to highlight valuable features and suppress useless ones. FL-CSE-ROIE is applied to the fashionable hybrid task cascade (HTC) model. Results on two open SAR ship detection dataset (SSDD) and high-resolution SAR images dataset (HRSID) confirm its effectiveness. Moreover, another two improvements to HTC are also proposed to enhance accuracy further: 1) the raw deconv is replaced with a content-aware reassembly of features block (CARAFEB) to enable larger receptive fields and 2) the raw$1\times1$conv for the mask information interaction is replaced with a global feature self-attention block (GFSAB) to enhance interaction benefits. Finally, FL-CSE-ROIE surpasses the other nine advanced models, better than the suboptimal model by 2.4%/2.3% detection average precision (AP) and 3.0%/2.5% segmentation AP on SSDD/HRSID.

Most of existing synthetic aperture radar (SAR) ship instance segmentation models do not achieve mask interaction or offer limited interaction performance. Besides, their multi-scale ship instance segmentation performance is moderate especially for small ships. To solve these problems, we propose a mask attention interaction and scale enhancement network (MAI-SE-Net) for SAR ship instance segmentation. MAI uses an atrous spatial pyramid pooling (ASPP) to gain multi-resolution feature responses, a non-local block (NLB) to model long-range spatial dependencies, and a concatenation shuffle attention block (CSAB) to improve interaction benefits. SE uses a content-aware reassembly of features block (CARAFEB) to generate an extra pyramid bottom-level to boost small ship performance, a feature balance operation (FBO) to improve scale feature description, and a global context block (GCB) to refine features. Experimental results on two public SSDD and HRSID datasets reveal that MAI-SE-Net outperforms the other nine competitive models, better than the suboptimal model by 4.7% detection AP and 3.4% segmentation AP on SSDD and by 3.0% detection AP and 2.4% segmentation AP on HRSID.

Huge imbalance of different scenes’ sample numbers seriously reduces synthetic aperture radar (SAR) ship detection accuracy. Thus, to solve this problem, this letter proposes a balance scene learning mechanism (BSLM) for offshore and inshore ship detection in SAR images. BSLM involves three steps: 1) based on unsupervised representation learning, a generative adversarial network (GAN) is used to extract the scene features of SAR images; 2) using these features, a scene binary cluster (offshore/inshore) is conducted by${K}$-means; and 3) finally, the small cluster’s samples (inshore) are augmented via replication, rotation transformation or noise addition to balance another big cluster (offshore), so as to eliminate scene learning bias and obtain balanced learning representation ability that can enhance learning benefits and improve detection accuracy. This letter applies BSLM to four widely used and open-sourced deep learning detectors, i.e., faster regions-convolutional neural network (Faster R-CNN), Cascade R-CNN, single shot multibox detector (SSD), and RetinaNet, to verify its effectiveness. Experimental results on the open SAR ship detection data set (SSDD) reveal that BSLM can greatly improve detection accuracy, especially for more complex inshore scenes.

The success of deep learning in synthetic aperture radar (SAR) automatic target recognition (ATR) relies on a large number of labeled samples; however, there are often wrong (noisy) labels in a large-scale dataset. In this article, we propose a loss curve-fitting-based method, which can identify the noisy labels and train the classification network effectively. We propose to model label noise by unsupervised clustering via fitting loss curve to identify whether the sample’s label is clean or noisy. Then, we train the network using augmented samples with clean labels to correct noisy labels further. The experiments on the moving and stationary target acquisition and recognition (MSTAR) dataset prove that our proposed method can deal with the situation when training a network with different ratios of noisy labels and correct noisy labels effectively. When the noise ratio is small (40%) in the training dataset, our method can correct 97.9% of noisy labels and train the classification network with 98.8% classification accuracy. While the noise ratio is large (80%), our method can correct 78.1% of noisy labels and train the classification network with 79.6% classification accuracy.

Ship classification in synthetic aperture radar (SAR) images is a fundamental and significant step in ocean surveillance. Recently, with the rise of deep learning (DL), modern abstract features from convolutional neural networks (CNNs) have hugely improved SAR ship classification accuracy. However, most existing CNN-based SAR ship classifiers overly rely on abstract features, but uncritically abandon traditional mature hand-crafted features, which may incur some challenges for further improving accuracy. Hence, this article proposes a novel DL network with histogram of oriented gradient (HOG) feature fusion (HOG-ShipCLSNet) for preferable SAR ship classification. In HOG-ShipCLSNet, four mechanisms are proposed to ensure superior classification accuracy, that is, 1) a multiscale classification mechanism (MS-CLS-Mechanism); 2) a global self-attention mechanism (GS-ATT-Mechanism); 3) a fully connected balance mechanism (FC-BAL-Mechanism); and 4) an HOG feature fusion mechanism (HOG-FF-Mechanism). We perform sufficient ablation studies to confirm the effectiveness of these four mechanisms. Finally, our experimental results on two open SAR ship datasets (OpenSARShip and FUSAR-Ship) jointly reveal that HOG-ShipCLSNet dramatically outperforms both modern CNN-based methods and traditional hand-crafted feature methods.

With the rise of Deep Learning (DL), numerous DL-based SAR ship detectors, represented by Faster R-CNN, is constantly breaking the record of detection accuracy. However, these detectors still face huge challenges in modeling the geometric transformation of shape-changeable ships, due to their used conventional convolution kernels whose structure is fixed. Therefore, to address this problem, we propose an improved Faster R-CNN by using deformable convolution kernels for SAR ship detection. We substitute some conventional shape-changeless convolution kernels in Faster R-CNN with deformable convolution ones that can adaptively learn additional 2-D offsets of the raw convolution kernels, to better model the geometric transformation of shape-changeable ships. Finally, the experimental results on the open SAR Ship Detection Dataset (SSDD) reveal that our improved Faster R-CNN achieves a 2.02% mean Average Precision (mAP) improvement than the raw Faster R-CNN.

Ship classification in Synthetic Aperture Radar (SAR) images is significant but its application based on Convolutional Neural Network (CNN) has not been adequately studied. Considering that there will be the loss of SAR ship spatial information as the network deepening in CNN, which is a great obstacle for the further improvement of algorithm accuracy. Thus, to deal with the problem, in this paper, a novel multi-scale CNN (MS-CNN) is proposed. MS-CNN can utilize the multi-scale features to enhance the feature expression ability by the following three steps, namely flattening, integrating and classifying. As a result, the experiments on the OpenSARShip dataset show that MS-CNN can increase the classification accuracy by 4.81% than benchmark network.

Ship classification in Synthetic Aperture Radar (SAR) images is a fundamental and important step in ocean surveillance. Recently, with the rise of Deep Learning (DL), Convolutional Neural Network (CNN)-based SAR ship classifiers have made a huge accuracy progress compared with traditional hand-crafted feature methods. However, existing most CNN-based classification models uncritically abandon traditional mature hand-crafted features, but excessively rely on abstract features extracted by deep networks, which possibly brings great challenges in further improving classification performance. Therefore, to address this problem, this paper proposes a Histogram of Oriented Gradient (HOG) feature fusion method to improve CNN-based SAR ship classification accuracy. Experimental results on the open SAR ship classification dataset OpenSARShip reveal that when combining HOG feature fusion, the classification accuracy can achieve a 7.64% improvement.

Existing most deep learning-based synthetic aperture radar (SAR) ship detectors have huge network scale and big model size. Thus, to solve these defects, we propose a lightweight SAR ship detector “ShipDeNet-20” with 20 convolution layers and <; 1 MB (0.82 MB) model size. We use fewer layers and kernels, and depthwise separable convolution (DS-Conv) to ensure ShipDeNet-20's lightweight attribute. Moreover, we also propose a feature fusion module (FF-Module), a feature enhance module (FE-Module), and a scale share feature pyramid module (SSFP-Module) to compensate for the raw ShipDeNet-20's accuracy loss. Experimental results on the open SAR ship detection data set (SSDD) reveal that the accuracy and speed of ShipDeNet-20 are both superior to the other nine state-of-the-art object detectors. Finally, detection results on another two wide-region SAR images show ShipDeNet-20's strong migration ability. ShipDeNet-20 is a novel SAR ship detector, built from scratch, lighter than others by tens even hundreds of times, helpful for real-time SAR application and future hardware transplantation.

With the rise of Artificial Intelligence (AI), many previous studies have already applied Deep Learning (DL) for ship detection from Synthetic Aperture Radar (SAR) imagery. However, these network scale and model size are both rather huge, leading to more computation costs. As a result, ship detection speed is bound to decline due to more computation costs, and FPGA/DSP transplantation also becomes more challenging coming from huge mode size. Therefore, to solve these problems, this paper proposes a novel lightweight deep learning network for SAR ship detection named ShipDeNet-18 (only 18 convolution layers). Essentially, fewer layers and fewer kernels jointly contribute to ShipDeNet-18's light-weight characteristic. In addition, to compensate for the severe detection accuracy's sacrifice, we also propose a Deep and Shallow Feature Fusion Module (DSFF-Module) and a Feature Pyramid Module (FP-Module), which can effectively improve its detection accuracy. Experimental results on the open SAR Ship Detection Dataset (SSDD) reveal that ShipDeNet-18's detection speed is largely superior to the other state-of-the-art detectors, meanwhile its detection accuracy is only slightly inferior to others. ShipDeNet-18 is a brand-new deep learning network built from scratch, more light-weight than the other detectors, with fewer parameters (228,246), lower computation costs (456,042 FLOPs), and smaller model size (1 MB). It is of great value in some real-time SAR application, and is also convenient for future hardware transplantation (FPGA/DSP).

This paper describes a new approach based on Transformation-Based Learning for extracting hierarchical decision rules. Genetic algorithms are adapted to establish the context environment for transformation operation and the transformation operation can lengthen the life cycle of “good” candidate rules. The experiments are conducted on iris, wine and glass datasets with a 10-fold cross validation setup. The results show that transformation operation can improve the precision of the classifier with a smaller number of rules and generations than hierarchical decision rules. The approach also works well in touching block extraction of Chinese handwritten text.

Desirable goals for grid task scheduling algorithms would shorten average delay, maximize system utilization and fulfill user constraints. In this work, an agent-based grid management infrastructure coupled with mean field annealing (MFA) scheduling algorithm has been proposed. An agent in grid utilizes a neural network algorithm to manage and schedule tasks. The Hopfield neural network is good at finding optimal solution with multi-constraints and can be fast to converge to the result. However, it is often trapped in a local minimum. Stochastic simulated annealing algorithm has an advantage in finding the optimal solution and escaping from the local minimum. Both significant characteristics of Hopfield neural network structure and stochastic simulated annealing algorithm are combined together to yield a mean field annealing scheme. A modified cooling procedure to accelerate reaching equilibrium for normalized mean field annealing has been applied to this scheme. The simulation results show that the scheduling algorithm of MFA works effectively.

Directional features are preferred in off-line Chinese character recognition due to their superior performance. This paper proposes an enhanced four plane feature (en-FPF) within a segmentation-free recognition framework. First, the directional planes are strengthened by replenishing salient pixels. Second, the method to count perpendicular strokes are renewed. In experiments of realistic Chinese handwriting recognition, the proposed enhancement yields desirable improvements of recognition rates, especially to punctuation marks and digits. Compared with four-orientation Gradient feature and Gabor feature, the superiority of en-FPF is also observed.

A segmentation-free strategy based on hidden Markov models (HMMs) is presented for offline recognition of unconstrained Chinese handwriting. As the first step, handwritten textlines are converted to observation sequence by sliding windows and character segmentation stage is avoided prior to recognition. Following that, embedded Baum-Welch algorithm is adopted to train character HMMs. Finally, best character string maximizing the a posteriori is located through Viterbi algorithm. Experiments are conducted on the HIT-MW database written by more than 780 writers. The results show: First, our baseline recognizer outperforms one segmentation-based OCR product with 35% relative improvement; second, more discriminative feature and compact representation, and state-tying technique to alleviate the data sparsity can enhance the recognizer with high confidence. The final recognizer has improved the performance by 10.77% than the baseline system.

This paper proposes a skew detection method for real Chinese handwritten documents. After analyzing the characteristics of Chinese characters, it utilizes the horizontal stroke histogram. Its accuracy, ability to increase the recall rate of text line separation, and CPU time consuming are investigated using 853 real Chinese handwritten documents. The results show that: 1) the method can identify 98.83% of the skew angles within one degree, with an improvement of 8.44% than Wigner-Ville distribution (WVD) method; 2) when incorporated into text line separation, the recall rate has an improvement of 2.54% than WVD method; 3) the method only consumes one-twentieth of WVD method on the same test environment.

Novel features and weak classifiers are proposed for face detection within the AdaBoost learning framework. Features are histograms computed from a set of spatial templates in filtered images. The filter banks consist of Intensity, Laplacian of Gaussian (Difference of Gaussians), and Gabor filters, aiming to capture spatial and frequency properties of faces at different scales and orientations. Features selected by AdaBoost learning, each of which corresponds to a histogram with a pair of filter and template, can thus be interpreted as boosted marginal distributions of faces. We fit the Gaussian distribution of each histogram feature only for positives (faces) in the sample set as the weak classifier. The results of the experiment demonstrate that classifiers with corresponding features are more powerful in describing the face pattern than haar-like rectangle features introduced by Viola and Jones.

BACKGROUND: It is one of the ultimate goals for modern biological research to fully elucidate the intricate interplays and the regulations of the molecular determinants that propel and characterize the progression of versatile life phenomena, to name a few, cell cycling, developmental biology, aging, and the progressive and recurrent pathogenesis of complex diseases. The vast amount of large-scale and genome-wide time-resolved data is becoming increasing available, which provides the golden opportunity to unravel the challenging reverse-engineering problem of time-delayed gene regulatory networks. RESULTS: In particular, this methodological paper aims to reconstruct regulatory networks from temporal gene expression data by using delayed correlations between genes, i.e., pairwise overlaps of expression levels shifted in time relative each other. We have thus developed a novel model-free computational toolbox termed TdGRN (Time-delayed Gene Regulatory Network) to address the underlying regulations of genes that can span any unit(s) of time intervals. This bioinformatics toolbox has provided a unified approach to uncovering time trends of gene regulations through decision analysis of the newly designed time-delayed gene expression matrix. We have applied the proposed method to yeast cell cycling and human HeLa cell cycling and have discovered most of the underlying time-delayed regulations that are supported by multiple lines of experimental evidence and that are remarkably consistent with the current knowledge on phase characteristics for the cell cyclings. CONCLUSION: We established a usable and powerful model-free approach to dissecting high-order dynamic trends of gene-gene interactions. We have carefully validated the proposed algorithm by applying it to two publicly available cell cycling datasets. In addition to uncovering the time trends of gene regulations for cell cycling, this unified approach can also be used to study the complex gene regulations related to the development, aging and progressive pathogenesis of a complex disease where potential dependences between different experiment units might occurs.

This paper presents novel likelihood estimation to be used for particle filter based object tracking. The likelihood estimation is built upon cascade object detector trained with Gentle AdaBoost (GAB), in order to capture the probability of existence of object. Two strategies are adopted to construct the likelihood functions: probability-intra-stage (PIS) corresponding to real output of each weak classifier in the same stage, and probability-outer-stage (POS) corresponding to the depth reached in the cascade detector. Five kinds of likelihood functions are thus proposed based on the trained GAB detector. Our experiment shows the likelihood functions are able to characterize probabilistically the existence of object accurately, having much higher confidence value in object regions than that in background, and that the integral strategy of PIS and POS is the best choice.

BACKGROUND: Development of robust and efficient methods for analyzing and interpreting high dimension gene expression profiles continues to be a focus in computational biology. The accumulated experiment evidence supports the assumption that genes express and perform their functions in modular fashions in cells. Therefore, there is an open space for development of the timely and relevant computational algorithms that use robust functional expression profiles towards precise classification of complex human diseases at the modular level. RESULTS: Inspired by the insight that genes act as a module to carry out a highly integrated cellular function, we thus define a low dimension functional expression profile for data reduction. After annotating each individual gene to functional categories defined in a proper gene function classification system such as Gene Ontology applied in this study, we identify those functional categories enriched with differentially expressed genes. For each functional category or functional module, we compute a summary measure (s) for the raw expression values of the annotated genes to capture the overall activity level of the module. In this way, we can treat the gene expressions within a functional module as an integrative data point to replace the multiple values of individual genes. We compare the classification performance of decision trees based on functional expression profiles with the conventional gene expression profiles using four publicly available datasets, which indicates that precise classification of tumour types and improved interpretation can be achieved with the reduced functional expression profiles. CONCLUSION: This modular approach is demonstrated to be a powerful alternative approach to analyzing high dimension microarray data and is robust to high measurement noise and intrinsic biological variance inherent in microarray data. Furthermore, efficient integration with current biological knowledge has facilitated the interpretation of the underlying molecular mechanisms for complex human diseases at the modular level.

A novel energy-minimizing model is proposed in this paper for motion detection and object tracking. In our model, the logarithmic image is employed to compensate for illumination change. Another feature of our approach is that the boundary force is no longer necessary. For solving the corresponding curve evolution equation efficiently, a local level set algorithm is adopted. When constructing the narrow band and resetting the level set function to the signed distance function, it is not needed to explicitly label the contour points during the evolution of contours. The local level set algorithm is based on partial differential equations (PDEs), which leads to a simple, flexible and stable scheme. This paper also proposes an appropriate stopping criterion for the level set algorithm without a need of explicitly extracting the locations of the evolving curve.