Search

Learning effective feature representations with deep convolutional neural network (CNN) and metric learning methods to distinguish pedestrians is the key to the success of recent advances in person re-identification (re-ID) tasks. However, the features provided by the common CNN network are not strong enough to distinguish between similar subjects because these common features describe only the scattered local patterns while neglecting the correlation and combinations between them. In fact, the high-order correlations of common features can be significant to the recognition of identities. In this paper, to tackle this problem, we design a flexible high-order block (HOB) module and a scheme of deep metric learning to produce the high-order representations of deep features for the re-identification of pedestrians. Extensive experiments prove the superiority of our proposed HOB module for person re-ID issue. On three large-scale datasets, including Market-1501, DukeMTMC-ReID, and CUHK03-NP, the HOB-net method achieves the competitive results with the state-of-the-arts, particularly in the mAP.

The salient target detection aims to use models or algorithms to detect and segment the salient target regions in the image. As an image preprocessing step, The salient target detection has been widely researched and applied in computer vision fields such as image segmentation, image compression, object recognition, and image retrieval. In the face of complex environments, such as poor lighting conditions and cluttered backgrounds, saliency detection is still challenging. RGB-T complementary fusion for saliency detection has become a hot research direction to solve the above problems. Therefore, this paper makes full use of the advantages of RGB and thermal infrared images to complement each other, and proposes an end-to-end salient target detection network. To this end, the VGG16 backbone network is first used for rough feature extraction, and then the attention mechanism module is used to enhance and merge the high-level features of RGB and thermal infrared images to enrich semantic information. Then the DFPN module is designed to accumulate high-level features to obtain high-level semantic information, and at the same time, the high-level features integrate the semantic information of thermal infrared images. Then we designed the SIPM self-interactive pooling module to further refine the features so that it can adaptively handle multi-scale targets. Finally, we compared the proposed method with several other advanced methods on several public datasets, and the experimental results proved the effectiveness of our method.

Stereo depth estimation based on learning with CNNs has become the dominant paradigm due to their significant leap in performance. However, since the inconsistency of the distribution between the training data (synthetic image) and the actual scene, the generalization ability is insufficient for deployment. Recently, the work MADNet with modular adaptation was proposed to tackle this issue on the fly. However, it still runs slowly with not-so-ideal performance. This paper introduces a “harmless training or inference” (HToI) framework to improve their work in both convergence and prediction speed. Firstly, in the actual scenario, the images arrive in sequence without ground-truth. Online adaptation to new target domains is only through self-supervised loss with errors (e.g., occluded areas). This paper introduces a confidence-weighted map (CWM) to mask the incorrect loss. The confidence map will guide the model to use the correct loss for online adaptation to obtain better accuracy. Secondly, this paper proposes a supervisor, implemented by Mann-Kendall trend detection on loss (MKL), to monitor the performance changes during the adaptation. According to the MKL, the model can autonomously determine whether to keep harmless training. If not training (i.e., only inference), the model will only predict the disparity, significantly reducing the running time. Finally, experiments on real-world stereo datasets confirm the effectiveness of CWM and MKL, which can make the model adapt to the scene more stably and accurately.

Three-dimensional depth perception using structured light has been widely researched and developed due to its robustness, high precision, and density. An open research issue exists with current structured light technology to obtain depth maps of scenes with objects in different motion states, while ensuring high accuracy and resolution. In this paper, a new algorithm for depth acquisition of multi-state targets is designed. A novel projection pattern is designed based on sinusoidal, speckle and line patterns. The foreground targets are separated according to the background segmentation technology, and regions with different motion states are divided into static, slow motion or fast motion regions. Then, the different motion states are used to select different pattern decoding schemes. In the experiments, the depth of multi-state objects are successfully calculated by our technique.

A large class of Person Re-identification (ReID) approaches identify pedestrians with the TriHard loss. Though the TriHard loss is a robust ReID method, pose variance and viewpoint in pedestrians constrain the performance. To address this problem, we introduce a spatial transformer network (STN) to align pedestrians. Then, we illustrate the generality of the STN module in pose variance problem through the evaluations on feature representation network (FRN) like VGG, ResNet and DenseNet architectures respectively. Furthermore, based on the evaluation results, we propose a robust and high-performance ReID model which consists of the STN module, DenseNet backbone and TriHard loss. And finally, we prove that our ReID model is whole differentiable by formula derivation, therefore achieving an end-to-end high-performance ReID system. The experiments show that our ReID system outperforms the state-of-art methods on Market-1501, DukeMTMC-reID and CUHK03 datasets.

Convolutional autoencoders are making a significant impact on computer vision and signal processing communities. In this work, a convolutional autoencoder denoising method is proposed to restore the corrupted laser stripe images of the depth sensor, which directly reduces the external noise of the depth sensor so as to increase its accuracy. To reduce the amount of training data and avoid overfitting, a patch size of the laser stripe image is determined, on the basis of which a small-scale dataset called Laser Stripe Image Patch (LSIP) is created. Also, a 14-layers convolutional autoencoder is constructed to reduce the noise of the image patches, which can learn the most salient features on the LSIP dataset. Moreover, the trained convolutional autoencoder is applied to an omnidirectional structured light system. Experimental results demonstrate that the proposed method obtains useful features and superior performance both visually and quantitatively on denoising tasks, and significantly improves the accuracy of the structured light system. Keywords: Convolutional autoencoder, Deep learning, Depth perception, Image denoising, Laser stripe, Structured light

The omnidirectional ring structured light depth perception system can project a 360-degree structured light.However, due to the influence of ambient light, refraction and other noise factors, it is easy to cause false detection of structured light. Therefore, denoising plays an important role in the omnidirectional ring structured light depth perception system.Firstly, this paper uses the DCGAN network to generate more data sets due to the lack of data sets. Secondly, based on the above data sets, this paper uses the autoencoder network to denoise the ring structured light. Experiments show that by using a combination of DCGAN and autoencoder network for denoising, the structured light’s noise is lesstherefore more robust to the surrounding environment compared to other methods.

A depth perception method based on omnidirectional image and ring structured light is proposed. Firstly, a new omnidirectional ring structured light pattern (ORSL) was presented. ORSL can simplify feature point extraction and locate the scene position more robust and fast. Secondly, a new sensor is proposed, which including an omnidirectional camera, laser projector and control circuit system. Thirdly, measurement accuracy was improved on error correction methods. Finally, a pose estimation method is designed to compute the location and direction of the imaging system. The experimental results demonstrated that the proposed method can acquire omnidirectional depth information about large-scale scene. The error analysis of multiple sets of depth data show the effectiveness of the method.

To better understand task failures in cloud computing systems, the authors analyze failure frequency of tasks based on Google cluster dataset, and find some frequently failing tasks that suffer from long-term failures and repeated rescheduling, which are called killer tasks as they can be a big concern of cloud systems. Hence there is a need to analyze killer tasks thoroughly and recognize them precisely. In this article, the authors first investigate resource usage pattern of killer tasks and analyze rescheduling strategies of killer tasks in Google cluster to find that repeated rescheduling causes large amount of resource wasting. Based on the above observations, they then propose an online killer task recognition service to recognize killer tasks at the very early stage of their occurrence so as to avoid unnecessary resource wasting. The experiment results show that the proposed service performs a 93.6% accuracy in recognizing killer tasks with an 87% timing advance and 86.6% resource saving for the cloud system averagely.

As an important and challenging task for intelligent video surveillance systems, video anomaly detection is generally referred to as automatic recognition of video frames that contain abnormal targets, behavior or events. Although it has been widely applied in real scenes, anomaly detection remains a challenging task because of the vague definition of anomaly and the lack of the anomaly samples. Inspired by the widespread application of Generative Adversarial Network (GAN), we propose an end-to-end pipeline called NM-GAN which assembles an encode-decoder reconstruction network and a CNN-based discrimination network in a GAN-like architecture. The generalization ability of the reconstruction network is properly modulated via the adversarial learning around reconstruction error maps and noise maps. Meanwhile, the discrimination network is trained to distinguish anomaly samples from normal samples based on the reconstruction error maps. Finally, the output of the discrimination network is transferred to evaluate anomaly score of the input frame. The thorough proof-of-principle experiments and ablation tests on several popular datasets reveal that the proposed model enhance the generalization ability of the reconstruction network and the distinguishability of the discrimination network significantly. The comparison with the state-of-the-art shows that the proposed NM-GAN model outperforms most competing models in precision and stability.

High-precision and high-resolution depth perception based on structured light is widely used. This measurement system consists of projector and camera equipment. Depth sensing based on the structured light method, the calibration process of obtaining the intrinsic and extrinsic parameters of the measurement system is a necessary step. However, the traditional calibration method has certain limitations. In this paper, we propose a self-calibration method. We overcome the problems in existing calibration techniques, and can quickly obtain the intrinsic and extrinsic parameters of the random speckle structured light system in motion scene. We designed a random speckle matching algorithm and proposed a non-coplanar point acquisition scheme. An effective method for finding the correspondence points between dense point pattern views is introduced. Next, the self-calibration process is described, using non-coplanar points matching to calculate the parameters required by the random speckle structured light system in real time. The results obtained in this paper are shown in the experimental section and analyzed from both quantitative and qualitative aspects.

Bugs severely damage the reliability of open source software. In order to improve the reliability of open source software, bug tracking system is built to collect and manage bugs reported from users all over the world. When system failures occur, users investigate whether failures are induced by software bugs and then report bugs. However, it is usually difficult and time consuming to identify bugs from system failures. To accelerate bug reporting and reduce the time users spend on identifying bugs, we present BugIdentifier, an automatic bug identifying approach based on log mining. BugIdentifier combines Doc2Vec with Deep Neural Network (DNN) and treats bug identifying as a binary classification problem. Doc2Vec is adopted to train a log sequence embedding model that transforms log sequences into feature vectors, and then DNN is used to identify whether the log sequence is bug-induced or not. The results of our empirical evaluation show that our approach can automatically identify real-world bugs of Hadoop and OpenStack with the F1-score higher than 75%, specifically, old-version bugs of OpenStack can be identified with 97% F1-score, as a result, bug reporting can be accelerated correspondingly.

We propose an image-based face swapping algorithm, which can be used to replace the face in the reference image with the same facial shape and features as the input face. First, a face alignment is made based on a group of detected facial landmarks, so that the aligned input face and the reference face are consistent in size and posture. Secondly, an image warping algorithm based on triangulation is presented to adjust the reference face and its background according to the aligned input faces. In order to achieve more accurate face swapping, a face parsing algorithm is introduced to realize the accurate detection of the face-ROIs, and then the face-ROI in the reference image is replaced with the input face-ROI. Finally, a Poisson image editing algorithm is adopted to realize the boundary processing and color correction between the replacement region and the original background, and then the final face swapping result is obtained. In the experiments, we compare our method with other face swapping algorithms and make a qualitative and quantitative analysis to evaluate the reality and the fidelity of the replaced face. The analysis results show that our method has some advantages in the overall performance of swapping effect.

Graph-based salient object detection methods have gained more and more attention recently. However, existing works fail to separate effectively salient object and background in some challenging scenes. Inspired by this observation, we propose an effective salient object detection method based on a novel boundary-guided graph structure. More specifically, the input image is firstly segmented into a series of superpixels. Then we integrate two prior cues to generate the coarse saliency map, a novel weighting mechanism is proposed to balance the proportion of two prior cues according to their performance. Secondly, we propose a novel boundary-guided graph structure to explore deeply the intrinsic relevance between superpixels. Based on the proposed graph structure, an iterative propagation mechanism is constructed to refine the coarse saliency map. Experimental results on four datasets show adequately the superiority of the proposed method than other state-of-the-art methods.

A depth perception method combining omnidirectional images and encoding structured light was proposed. First, a new structured light pattern was presented by using monochromatic light. The primitive of the pattern consists of four-direction sand clock-like (FDSC) image. FDSC can provide more robust and accurate position compared with conventional pattern primitive. Second, on the basis of multiple reference planes, a calibration method of projector was proposed to significantly simplify projector calibration in the constructed omnidirectional imaging system. Third, a depth point cloud matching algorithm based on the principle of prior constraint iterative closest point under mobile condition was proposed to avoid the effect of occlusion. The experimental results demonstrated that the proposed method can acquire omnidirectional depth information about large-scale scenes. The error analysis of 16 groups of depth data reported a maximum measuring error of 0.53 mm and an average measuring error of 0.25 mm.

In this paper, a new saliency detection model is proposed based on a space-to-frequency transformation. Firstly, the equivalence of spatial filtering and spectral modulation is demonstrated to explain the intrinsic mechanism of typical frequency-based saliency models. Then a novel frequency-based saliency model is presented based on the Fourier Transformation of multiple spatial Gabor filters. Besides, a new saliency measurement is proposed to implement the competition between saliency maps at multiple scales and the fusion of color channels. In experiments, we use a set of typical psychological patterns and four popular human fixation datasets to test and evaluate the proposed model. In addition, a new energy-based criterion is proposed to evaluate the performance of our model and is compared with five traditional saliency metrics for validation. Experimental results show that our model outperforms most of the competing models in salient object detection and human fixation prediction. HighlightsSome frequency-based saliency models can be transformed into the space domain.The frequency-based model derived from the transformation of multiple Gabor filters outperforms most of competing models.The SWE-based scale competition is a better scheme for scale combination.ARI is a better saliency metric than other AUC-based metrics.

This paper presents a depth perception system based on one-shot coding structured light method. The designed monochromatic pattern is robust against occlusion and sharp deformation, suitable for dynamic scene, and less sensitive for color interference, with symbol density spectrum (SDS) rule, legitimately increasing number of feature points inside element which further enhancing resolution. For the decoding stage, firstly, the proposed adaptive-symmetry-template (ASD) can robustly detect feature points, secondly, in virtue of transfer learning, using DenseNet to identify element label, thirdly, graph topological structure and look-up-table (LUT) help reducing decoding time. The experiments show that the encoding method increases resolution fairly a lot and for several complex objects, the decoding method can achieve high accuracy and keep robustness.

With the development of information technology and the popularization of monitoring network, how to quickly and automatically detect abnormal behaviors in surveillance video is becoming more and more important for public security and smart city. The emergence of deep learning has greatly promoted the development of anomaly detection and much remarkable work has been presented on this topic. However, the existing approaches for anomaly detection generally encounter problems such as insufficient utilization of motion patterns and instability on different datasets. To improve the performance of anomaly detection in surveillance video, we propose a framework based on bidirectional prediction, which predicts the same target frame by the forward and the backward prediction subnetworks, respectively. Then the loss function is constructed based on the real target frame and its bidirectional prediction frame. Furthermore, we also propose an anomaly score estimation method based on the sliding window scheme which focuses on the foregrounds of the prediction error map. The comparison with the state-of-the-art shows that the proposed model outperforms most competing models on different video surveillance datasets.

Macular disorders, such as age-related macular degeneration (AMD) and diabetic macular edema (DME), have plagued humans for many years, yet the traditional recognition methods are often complicated, time-consuming and laborious. Therefore, this paper proposes an effective recombined residual convolutional neural network (CNN) with less computation to recognize AMD, DME and the normal of SD-OCT images. This paper adopts the18-layer Residual CNN as the basis and removes some of convolutional layers in the convolutional groups which extract low-level or high-level features, while add them to the middle part of the CNN, to enhance the ability of extracting mid-level features. This paper also adjusts the kernel size to change the visual receptive fields during convolution calculations. Experiments show that the recombined residual CNN with the kernel size of 3×3 shows better performance than the original one and recently proposed methods on the same dataset, with the highest overall accuracy up to 90%. Moreover, in terms of feature extraction complexity and time consumption, the recombined residual CNN is also more dominant.

Commodity display is very important in online shopping. In this paper, a method of reconstructing the commodity 3D model is proposed. Firstly, this paper accepts point clouds of a commodity from different angles. Secondly, before initial alignment, this paper compresses point clouds based on octree. Thirdly, this paper aligns all point clouds. At last, we reconstruct the commodity surface. In this step, a point cloud smoothing algorithm based on MLS and a Delaunay growth projection algorithm are proposed. Experimental results demonstrate that the computation time consumed by our improved registration algorithm is only one sixth of that of traditional one. And the quality only decreases slightly. The commodity 3D model has been successfully and efficiently reconstructed.

Faced with the problem of high cost and low accuracy of 3D modeling in industry, this paper build a low-cost 3D data measurement system with a self-made line laser scanner and improves the algorithm of extracting center of light bar. First, the self-made line laser scanner is used to collect the image information of the laser line on the surface of the measured object. Secondly, the center position of the line laser image is calculated by the improved gray centroid algorithm, and the depth information of the target is obtained based on the triangulation principle. Finally, the conversion from the depth information to the three-dimensional coordinate in space is completed through establishing the Cartesian coordinate system, and the three-dimensional coordinates are visualized as the point cloud model. According to the experimental data, the measurement error of the 3D data measurement system designed in this paper is 5mm in the short distance (less than 1m). And the improved gray centroid algorithm proposed is more accurate, faster and better robust, which will have a broader application prospect.

Superpixel segmentation is a widely used preprocessing method in computer vision, but its performance is unsatisfactory for color images in cluttered indoor environments. In this work, a superpixel method named weighted coplanar feature clustering (WCFC) is proposed, which produces full coverage of superpixels in RGB-depth (RGBD) images of indoor scenes. Basically, a linear iterative clustering is adopted based on a cluster criterion that measures the color similarity, space proximity and geometric resemblance between pixels. However, to avoid the adverse impact of RGBD image flaws and to make full use of the depth information, WCFC first preprocesses the raw depth maps with an inpainting algorithm called a Cross-Bilateral Filter. Second, a coplanar feature is extracted from the refined RGBD image to represent the geometric similarities between pixels. Third, combined with the colors and positions of the pixels, the coplanar feature constructs the feature vector of the clustering method, thus, the distance measure, as the cluster criterion, is computed by normalizing the feature vectors. Finally, in order to extract the features of the RGBD image dynamically, a content-adaptive weight is introduced as a coefficient of the coplanar feature, which strikes a balance between the coplanar feature and other features. Experiments performed on the New York University (NYU) Depth V2 dataset demonstrate that WCFC outperforms the available state-of-the-art methods in terms of accuracy of superpixel segmentation, while maintaining a high speed.

This paper aims to study the construction of 3D temperature distribution reconstruction system based on depth and thermal infrared information. Initially, a traditional calibration method cannot be directly used, because the depth and thermal infrared camera is not sensitive to the color calibration board. Therefore, this paper aims to design a depth and thermal infrared camera calibration board to complete the calibration of the depth and thermal infrared camera. Meanwhile a local feature descriptors in thermal and depth images is proposed. The belief propagation matching algorithm is also investigated based on the space affine transformation matching and local feature matching. The 3D temperature distribution model is built based on the matching of 3D point cloud and 2D thermal infrared information. Experimental results show that the method can accurately construct the 3D temperature distribution model, and has strong robustness.

This paper presents a novel 3D depth information extraction method without calibration. Firstly, this paper develops an omni-directional 3D camera system, which consists of a CCD camera, hyperbolic mirror, infrared laser diodes and diffractive of element (DOE). Secondly, a depth measurement model is proposed to obtain the 3D depth information. Finally, in order to calculate the speckle shift accurately between the reference image and the object image, a dot matrix pattern and sequence coding algorithm are designed to find the corresponding speckles in the two images. Experimental results show that the reconstructed depth data have a good correlation with the actual distance. The accuracy of the data is also found to be influenced by the distance between the object and the camera. A novel omni-directional 3D camera framework is proposed.The omni-directional camera and infrared dot matrix structured light can be merged.The model can reveal the relation between the object depth and its pixel offset.We can obtain the pixel offset with the sequence coding algorithm.

This paper aims to study the projector calibration algorithm in omnidirectional structured light (OSL). The traditional projector calibration method can not directly be used in omnidirectional system, because the projector is perpendicular to the omnidirectional camera in our experiment. Therefor, we design a complete algorithm for the calibration of omnidirectional structured light. Firstly, a calibration plane is applied. And a checkerboard calibration board are placed on that and the checkerboard pattern projected from the projector onto that. Secondly, the equation of the calibration plane are computed based on the extrinsic parameters of the calibration board. Thirdly, the corners of the projected pattern are detected in the image captured by omnidirectional camera. Lastly, 3D projected points for each projected corner are obtained based on the ray-plane intersection. We designed a complete set of OSL calibration toolbox based on the proposed methods in Matlab. The proposed method and toolbox in Matlab have been shown to be accurate and easyto-use in projector calibration.

Weld seam tracking technique was a hot topic of welding automation study. In this paper, depth information extraction of the weld seam was developed based on binocular stereo vision. Firstly, the binocular camera was calibrated. After acquiring the intrinsic and extrinsic parameters of both left and right camera, the acquired sample images were corrected using epipolar constraint method. Then, disparity map was acquired utilizing stereo matching method. Finally, depth information of weld seam was extracted on triangulation method. In the experiments, we compared the binocular stereo vision and line structured light method. The experimental results indicated that the visual information of binocular camera was so rich and can get more 3D shape information of the weldments, which were advantageous to weld large weldments. The measurement accuracy of linear structured light was high.

This paper aims to study the construction of 3D temperature distribution reconstruction system based on binocular vision technology. Initially, a traditional calibration method cannot be directly used, because the thermal infrared camera is only sensitive to temperature. Therefore, the thermal infrared camera is calibrated separately. Belief propagation algorithm is also investigated and its smooth model is improved in terms of stereo matching to optimize mismatching rate. Finally, the 3D temperature distribution model is built based on the matching of 3D point cloud and 2D thermal infrared information. Experimental results show that the method can accurately construct the 3D temperature distribution model and has strong robustness.

Tobacco is widely planted worldwide as an important economic crop. The differences of planting environment and growth status in different regions lead to different levels of tobacco quality, while color is one of the important indexes to evaluate the quality grade of tobacco leaves. Even for those that are planted in the same area, colors of different grades of tobacco leaves vary greatly, so the color of tobacco leaves is often used as the main evaluating index in tobacco leaves grading. In this paper color extraction of standard tobacco leaves based on HSI color space was thoroughly studied. H, S and I component would be quantified and extracted by using color histogram, then the average value corresponding to every color component was calculated. After extraction and calculation on the same level of standard tobacco leaves, the ranges of three color components could be obtained in HSI color space by using statistical method, and provide data information for tobacco leaves grading.

Depth measurement is a challenging problem in computer vision research. In this study, we first design a new grid pattern and develop a sequence coding and decoding algorithm to process the pattern. Second, we propose a linear fitting algorithm to derive the linear relationship between the object depth and pixel shift. Third, we obtain depth information on an object based on this linear relationship. Moreover, 3D reconstruction is implemented based on Delaunay triangulation algorithm. Finally, we utilize the regularity of the error curves to correct the system errors and improve the measurement accuracy. The experimental results show that the accuracy of depth measurement is related to the step length of moving object.

Lung vessels often interfere with the detection of lung nodules. In this paper, a novel computer-aided lung nodule detection scheme on vessel segmentation is proposed. This paper describes an active contour model which can combine im- age region mean gray value and image edge energy. It is used to segment and remove lung vessels. A selective shape filter based on Hessian Matrix is used to detect suspicious nodules and remove omitted lung vessels. This paper extracts density, shape and position features of suspicious nodules, and uses a Rule-Based Classification (RBC) method to identify true posi- tive nodules. In the experiment results, the detection sensitivity is about 90% and FP is 1/scan.

Face Feature Point Detection (FFPD) is a significant and interesting topic in many related areas of face recognition. A new face mesh model is presented in this paper to realize the FFPD in a fast and training-free approach. At first, an automatic face mesh initialization algorithm based on the detection of face, eye-pairs and mouth is proposed. Second, a local operator referred as Edge Attractor and an adaptation algorithm based on global topological constraint are proposed to locate feature points at boundaries of facial features while maintain the global shape of the face mesh. At last, an iterative algorithm is developed to optimize the evaluation function of the face mesh. Experimental results showed that the proposed face mesh model is fast, robust, and accurate in detecting face feature points.

This paper studies the construction of 3D temperature distribution reconstruction system based on the binocular visual technology. First of all, since the thermal infrared camera is only sensitive to temperature, the traditional calibration method cannot be directly used in it. Therefore, the paper designs a thermal infrared camera calibration board to complete the calibration of the thermal infrared camera. Then, in terms of stereo matching, it further studies the traditional belief propagation algorithm and improves its smooth model, so as to optimize the mismatching rate greatly. Finally, in terms of the 3D model construction, it constructs the corresponding relationship between the points on the 3D model and the pixels on the temperature images based on the known thermal infrared camera and the space position information of the visible light camera and maps the temperature information to the 3D model. Experimental results show that the method can accurately construct the 3D temperature distribution model and the model has strong robustness.

In this paper, an accurate two-stage deep learning method is proposed to detect vulnerable plaques in ultrasonic images of cardiovascular. Firstly, a Fully Convonutional Neural Network (FCN) named U-Net is used to segment the original Intravascular Optical Coherence Tomography (IVOCT) cardiovascular images. We experiment on different threshold values to find the best threshold for removing noise and background in the original images. Secondly, a modified Faster RCNN is adopted to do precise detection. The modified Faster R-CNN utilize six-scale anchors (122,162,322,642,1282,2562) instead of the conventional one scale or three scale approaches. First, we present three problems in cardiovascular vulnerable plaque diagnosis, then we demonstrate how our method solve these problems. The proposed method in this paper apply deep convolutional neural networks to the whole diagnostic procedure. Test results show the Recall rate, Precision rate, IoU (Intersection-over-Union) rate and Total score are 0.94, 0.885, 0.913 and 0.913 respectively, higher than the 1st team of CCCV2017 Cardiovascular OCT Vulnerable Plaque Detection Challenge. AP of the designed Faster RCNN is 83.4%, higher than conventional approaches which use one-scale or three-scale anchors. These results demonstrate the superior performance of our proposed method and the power of deep learning approaches in diagnose cardiovascular vulnerable plaques.

Motivated by frequent failures in cloud computing systems, we analyze failure frequency and failure continuity of tasks from the Google cloud cluster, and find what we call killer tasks that suffer from frequent failures and repeated rescheduling. Killer tasks cause unnecessary resource wasting and significant increase of scheduling workloads, which can be a big concern in cloud systems. We aim to recognize killer tasks at the very early stage of their occurrence so that they can be addressed proactively instead of being rescheduled repeatedly, so as to promote reliability and save resources. To recognize killer tasks from a large amount of tasks in real time is really challenging. In this paper, we first investigate characteristics and behavior patterns of killer tasks and then develop two machine learning based methods, K-HUNTER and C-HUNTER, for online recognition of killer tasks. The empirical results show that our approach performs at 97% of precision in recognizing killer tasks with an 89% timing advance and 88% of resource saving for the cloud system on average.

Accurate and efficient face authentication method has been a hot research field for a long time. The main research attention of this paper was directed toward studying the face authentication based on RGB color image and depth image, which is captured by Kinect device. Firstly, the 121 face features was extracted based on Candide-3 face model. Secondly, this paper built a face mask using the characteristics value extracted from the depth image. Finally, this paper calculated the covariance of the characteristics value to analyze the similarity of different face images. Experiments results showed that the proposed method is fast, convenient and has good accuracy.

Benign and malignant lung nodules classification is an important task in the diagnosis of lung cancer. In this study, lung nodules are classified based on growth changes feature and registration technique. Firstly, this paper combine the global rigid registration with local elastic registration method, which can extract the growth changes of a region of interest. Secondly, the benign and malignant nodules are classified on a rule-based classifier. Experimental findings show that the proposed method can extract features automatically and yield accurate classification results.

Indoor scene understanding is important to service robots, and scene segmentation is part of this work. This paper has proposed a method of semantic cuboids to partition a single RGB-D image. Semantic cuboids is generated by combining semantic labels and cuboid. Semantic labels are assigned by Markov random field based on super-pixels. And cuboids are constructed by graph method in a scene. Experiments on real data show that the proposed framework obtains accurate semantic cuboids in indoor environments, and completes computational tasks in real time. This method executes faster and adds more semantic information to cuboid than linear matching approach of H Jiang. The results provide useful and precise information of indoor cuboids for service robots.

A new computer-aided detection (CAD) scheme for detecting lung nodules is proposed in this paper. Firstly, the lung region is segmented from the CT data using adaptive threshold algorithm etc; Secondly, building active contour model to segment and remove lung vessel accurately in the lung region; Next, suspicious nodules are detected and omitted renal vessel is filtered using a selective shape filter; Finally, nodule features are extracted and rule-based classifier is used to distinguish true or false positive nodules. Experiment results indicate that this scheme can help radiologist improve the diagnosis efficiency.

This paper had studied two methods for multichannel vibration signals compression, which were QDCT-based and QDFT-based method. In the experiment, the three signals data correspondent to orthogonal axes X, Y, Z were gotten by an acceleration sensor with three channels. The two methods could compress the multichannel vibration signal effectively, and also preserve the association for the signals of each channel, eliminate the noise and redundancy, extract the key features. The experiment results indicated its efficiency, and the comparative results indicated that the QDCT-based method was better than that of QDFT. Moreover that it provided a new method for multichannel signals to data compression and feature extraction.

Chan-Vese?s image segmentation method proposed based on level sets method is studied and improved in this paper. A new level set method has been presented, which combines the average values of global regions and local boundary information, and it is used for lung nodules detection for CT images. The experiment results verify that it is a robust way for the segmentation of lung body and the detection for lung nodules, the speed of calculation is faster than that of Chan-Vese?s. Moreover, the detection results are independed on the initial settings, so the manual mutuality can be reduced to the minimum, it is an effective method for CT image lung nodule detection automatically.

Although medical image segmentation is very important in the clinical application, how to obtain an accurate and quickly image segmentation is a difficulty to overcome because the distribution of gray is asymmetric and the number of the date is large. Basing the distance field, an algorithm of nonlinear image interpolation segmentation is presented. Algorithm adopts the cosine non-linear instead of the linear function to estimate the interpolation. In this process, the serial slices is segmented quickly and the result of the 3D reconstruction is relative accurate. The experiment shows that the algorithm can generate accurate result and can be successively applied in a 3D medical image visualization and reconstruction system.