Image and Graphics

In this paper, we propose a new pan-sharpening method by coupled unitary dictionary learning and clustered sparse representation. First, we randomly sample image patch pairs from the training images exclude the smooth patches, and divide these patch pairs into different groups by K-means clustering. Then, we learn sub-dictionaries offline from corresponding group patch pairs. Particularly, we use the principal component analysis (PCA) technique to learn sub-dictionaries. For a given LR MS patch, we adaptively select one sub-dictionary to reconstruct the HR MS patch online. Experiments show that the proposed method produces images with higher spectral resolution while maintaining the high-quality spatial resolution and gives better visual perception compared with the conventional methods.

Inpainting is an image processing method to automatically restore the lost information according to the existing image information. Inpainting has great application on restoration of the lost information for photographs, text removal of image, and recovery for the loss coding of image, etc. Image restoration based on partial differential equation (PDE) is an important repair technology. To overcome the shortcomings of the existing PDEs in repair process, such as false edge, incomplete interpolation information, a new PDE for image restoration based on image characteristics is proposed. The new PDE applies different diffusion mode for image pixels with the different characteristics, which can effectively protect the edges, angular points, and other important characteristics of the image during the repair process. The experimental results in both gray images and color images show that our method can obviously improve the image visual effect after inpainting compared with different traditional diffusion models.

In this paper a novel patch-based tracking algorithm is proposed by using two-stage multiple kernel learning. In the first stage, each object patch is represented with multiple features. Unlike simple feature combination, we utilize multiple kernel learning (MKL) method to obtain the optimal combination of multiple features and kernels, which assigns different weight to the features according to their discriminative power. In the second stage, we apply MKL to making full use of multiple patches of the target. This method can automatically distribute different weight to the object patches according to their importance, which improves the discriminative power of object patches as a whole. Within the Bayesian framework, we achieve object tracking by constructing a classifier, and the candidate with the maximum likelihood is chosen to be the target. Experiments demonstrate that the proposed tracking approach performs favorably against several state-of-the-art methods.

An effective method of pattern classification for dermoscopic images based on structure textons and Bag-of-Features (BoFs) model is proposed in this paper. Firstly, the pattern structures of images were enhanced. Secondly, images with obvious directivity were rotated to align their principal directions with horizontal axis, and Otsu method was used to obtain interesting regions. The intensity values of each pixel in the interesting region and its neighborhood composed patch vector. For each pattern, patch vectors of training images were clustered to generate K structure textons and a dictionary with 5 K elements was obtained. Then BoFs model was applied to obtain texton histograms for training and testing images respectively. Finally, a nearest neighbor classifier with chi-square distance was adopted to classify. The experimental results shows that our enhancement method is beneficial to pattern classification and correct classification rate achieves 91.87 %.

In view of the reality of facial expression animation and the efficiency of expression reconstruction, a novel method of real-time facial expression reconstruction is proposed. Our pipeline begins with the feature point capture of an actor’s face using a Kinect device. A simple face model has been constructed. 38 feature points for control are manually chosen. Then we track the face of an actor in real-time and reconstruct target model with two different deformation algorithms. Experimental results show that our method can reconstruct facial expression efficiency in low-cost. The facial expression of target model is realistic and synchronizes with the actor.

Human salience of pedestrians images is distinctive and has been shown importantly in person re-identification (or pedestrians identification) problem. Thus, how to obtain the salient area of pedestrian images is important for this salience based pedestrians identification problem. In this paper, we first show that this kind of salient area detection can be formulated as a kind of outlier detection problem, and then propose a novel unsupervised salience learning method using a local outlier-detection technique for person re-identification task. The main feature of the proposed salience computation method is that it exploits both distance and density information simultaneously. Experimental results on several datasets show the effectiveness of the proposed salience based person re-identification method.

This paper presents a phase unwrapping method of heterodyne three frequency non-equal step phase shift, to surmount disadvantages of needing dozens of images in 3-D measurement techniques of multi-frequency phase shift. The method selects three frequencies. Firstly, four-step phase-shifting algorithm is used to calculate wrapped phase and average intensity of the intermediate frequency. Then the wrapped phases of other frequencies are obtained utilizing two-step phase-shifting algorithm. Finally, the absolute phase is calculated using heterodyne method. In addition, the contour sine/cosine filter method is utilized to filter noise by the fringe orientation information. The 3-D points cloud of standard planar and calibration target are reconstructed experimentally. Then the planeness of standard planar is calculated. The centers of markers of calibration target are extracted to calculate distance between two adjacent circles. Experiments demonstrate that the proposed method reduces the number of fringe images, eliminates the effects of noise efficaciously.

This paper proposes a novel scalar contribution function for photon shooting to optimize the distribution of photons for scenes with complex lighting conditions. Normally, conventional particle tracing methods would become inefficient to render these scenes, where photons are poorly distributed due to occlusion. The new scalar contribution function combines the visual importance, the initial photon distribution and photon path visibility. Then adaptive Metropolis sampling on the function is implemented to shoot photons from the light source into the scene, effectively guiding more photons to critical area where the rendering error is prominent. Experimental results show that this approach can efficiently improve the photon distribution and produce images with less noise than state-of-the-art methods.

Palmvein recognition has emerged as a promising alternative for human recognition because of its uniqueness, permanence, acceptability, live body identification, and resistance to fraud. Palmvein image acquisition is the premise of palmvein recognition, the position and posture of the hand above the camera directly determines the quality of palmvein image. Palmvein image-capture device mostly through manual adjustment, cannot meet the demands for the practicability and productization. This paper proposes a simple and practical real-time locating method. A locating electronic and optical circuit is designed to capture an image with four light-spots, and a locating algorithm is constructed to detect the appropriate position and posture of hands. The experimental results illustrate that the capture device with the proposed approach can realize palmvein image acquisition automatically and quickly as well as to guarantee the validity and consistency of the acquired palmvein images.

To solve the problems such as low matching precision, high algorithm complexity and poor real-time in real-time panoramic image mosaic, this paper makes full use of the significance and parallelism of Harris corners as well as the invariance of feature describing methods in light intensity changes, translation and rotation, and proposes a real-time panoramic image mosaic algorithm which uses Harris corner detection and is logically implemented on FPGA. According to the basic features of classical algorithms and the properties of FPGA, several modules like feature point extraction, description and matching are optimized based on the logical implementation of FPGA. The optimized system realizes the high-precision matching in real time. The new algorithm collects images of 256 × 256 pixels by CCD camera on Spartan-6 hardware platform of Xilinx. After going through the algorithm, the mosaic images will finally be output on HD display in the form of DVI. The results show that the new algorithm based on FPGA has high precision, good real-time and robustness.

A real-time robust video stabilization system is proposed. Firstly, SIFT feature points are extracted and matched between the reference frame and current frame, and then global motion parameters are obtained by fitting the feature matches with random sample consensus algorithm. Secondly, multiple evaluation criteria, i.e., global motion parameters and location errors of corresponding feature matches, are fused by empirical mode decomposition to smooth global motion for obtaining correction vector. Thirdly, motion compensation is applied to the current frame by using correction vector. Finally, stabilized video is obtained after each frame is completed by combining the texture synthesis method and the spatio-temporal information of video. By comparing the jittered video and stabilized video, the experimental results demonstrate the system can increase the average peak signal-to-noise ratio around 7.2 dB, the identification ability and perceptive comfort on video content.

Absorption, scattering, and color distortion are three major issues in underwater optical imaging. In this paper, we propose a novel underwater imaging model that compensates for the attenuation discrepancy along the propagation path. In addition, we develop a fast weighted guided filtering algorithm for enhancing underwater optical transmission map. The final enhanced images are characterized by a reduced noised level, better exposure in dark regions, and improved global contrast, by which the finest details and edges are enhanced significantly. Our experiments show that with the use of our proposed algorithm, the peak signal-to-noise ratio is improved by at least 2 dB compared to existing state-of-the-art methods. The structural similarity index is improved by about 0.01.

Structural variations (SVs) are one of the genetic markers in the human genome and detecting them by using ultra high-throughput genome sequencing techniques has vital significance for genetic and evolutionary studies. In recent decades, bioinformatics techniques based on next-generation sequencing (NGS) have become a research focus owing to its high resolution and accuracy. Moreover, NGS devices are becoming cheaper. In this survey, we will summarize current methods based on next-generation sequencing algorithms for SVs detection and discuss the impacts of them. We also analyze the problems and give an outlook for the future research directions.

The three-dimensional interaction has been widely used as a natural and direct way in Human-Computer Interaction (HCI). In this paper, we propose a novel 3D interaction method by recognizing Chinese character written in the air. Firstly, the moving trajectory of fingertip is precisely captured using the Leap Motion Controller. Then, we describe the trajectory by combining the directional feature and direction-change feature. We construct a dataset called IAHCC-UCAS2014, which contains 3755 classes of Chinese characters and each character class has 65 samples. In the evaluation experiments, the proposed method shows promising recognition performance with little increase in computational cost.

We prove that the angle between any two Minkowski-reduced basis vectors is more than

$${\pi }/{3};$$

if the orthogonal defect of 3-dimension lattice is less than

$${2}/{\sqrt{3}},$$

the Minkowski-reduced basis of the lattice is

$${\pi }/{3}$$

-orthogonal; if a weakly

$$\theta $$

-orthogonal basis for a lattice with

$$\theta \geqslant {\pi }/{3}$$

has been ordered by the Euclidean norm of the vectors, and the minimum length ratio maximum length is more than

$$2\cos \theta ,$$

the basis is Minkowski reduced. We improve an algorithm used in JPEG CHEst by changing it from heuristic one to deterministic one, furthermore we add a constraint to reduce the number of unimodular matrix that need to determine.

abstract

environment.

Recently, no-reference image quality assessment has been followed with interest by researchers, but no-reference quality assessment of foggy images is rarely reported. This paper proposes a no-reference quality assessment of foggy images based on codebooks. Proposed method aims to be consistent with human subjective perception. The Technical roadmap of the method is from feature extraction to quality metric model. The features need to reflect characteristics of foggy images exactly. Then codebook is built by the features and used to acquire feature vectors of training images by encoding. At last, regression is introduced to quality model construction by feature vectors and subjective ratings. The method is tested in simulation library of foggy images. Results show that Pearson Linear Correlation Coefficient (PLCC) and Spearman rank Order Correlation Coefficient (SROCC) are both above 0.99. And compared with state-of-art algorithms, our method perceives higher performance, and it can be a good predictor of subjective perception of foggy images.

This paper presents a method of the vehicle type classification based on spatial pyramid representation and BP neural network. We extract feature vectors of each vehicle image by using the spatial pyramid representation method. By this way, we can use different size of pictures instead of changing the picture into a fixed size avoiding the deformation of the target images when cropping or warping and so on. We choose BP neural network to train our classifier and have a good performance on car, bus and truck classification.

One of the major research directions in robotic vision focuses on SLAM using RGB-D sensors. The information can be used for decision making of robots and other areas that require precise position as a feature. This paper presents a novel algorithm to calibrate the RGB-D sensors for service robots SLAM. The distortions of the RGB and depth images are calibrated before the sensor is used as a measuring device for robot navigation. The calibration procedure includes the correction of the RGB and depth image as well as alignment of the RGB lens with the depth lens. The key advances in this paper are: a new method for RDB-D sensors calibration, and use of a depth distortion correcting model to help improve measurement precision. We experimentally verify our algorithm using varies of methods. The results show that, typically, our approach provides accurate calibration and the RGB-D sensors could provide reliable measurement information for robots navigating in unknown environments.

As for vision based road detection, most of color based methods use a center-lower region as a “safe” road region to model road appearance. However, this region heavily relies on the pose of ego-vehicle. Color models trained by using samples from this region often yield biased results when some non-road regions are included. In this paper, we proposed a novel color based road detection method which can overcome this problem. It is based on an image boundary prior, which infers a road region by measuring the extent of the region connecting to the bottom boundary of an image. This prior is more robust than the center-lower prior. Moreover, we use illumination invariance color space for the distance metric of two neighboring regions in order to make our approach robust to shadows. Experiments demonstrate that the proposed method is superior to both the Gaussian mixture model based method and illumination invariance based method.

Lane marking detection is part of most advanced driver assistance systems (ADAS) as an important component of computer vision. This paper presents a lane detection system based on a novel lane feature extraction approach. The robustness and real-time of algorithm enable different configurations of embedded solutions. The system is divided into three phases. Firstly, using the Prewitt operator we can get the rich useful details and using Shen Jun operator we can get step edge, on the other hand Shen Jun operator is the best filter to detect the symmetrical markings according to the maximum signal noise ratio (SNR) criterion. So we introduce the best compromise method between noise smoothing and edge locating that combining the Prewitt operator with Shen Jun operator to extract lane markings. Then a fast Hough transform based on image pyramid is applied to get the lane lines. The posterior algorithm of reasonably refining the Lane lines angle is introduced to correct to error caused by Hough transform. Finally, robust detection of vehicle’s departure warning is also discussed. Experiment results on real road will be presented to prove the robustness and effectiveness of the proposed lane detection algorithm.

In traditional contour tracker, object’s location and shape are usually bound together to form the system state. Such approaches suffer from the problem that most sampled states cannot match the object’s boundary exactly when the boundary cannot be captured by the shape model. To overcome such drawbacks, Constrained Separate Tracking of Location and Shape (CSTLS) is proposed. In CSTLS, location and shape are tracked by separate tracker, L-Tracker and S-Tracker, with the constraints enforced by the global contour tracking. The likelihood measurement for each sample in L-Tracker/S-Tracker is calculated by taking multiple shape/location hypotheses into consideration, which help to improve the robustness of tracking. The relationships of L-Tracker and S-Tracker with original problem are established under Sequential Mean Field Monte Carlo method. Experiments demonstrate the effectiveness of the CSTLS.

Face image super resolution, also referred to as face hallucination, is aiming to estimate the high-resolution (HR) face image from its low-resolution (LR) version. In this paper, a novel two-layer face hallucination method is proposed. Different from the previous SR methods, by applying global similarity selecting, the proposed approach can narrow the scope of samples and boost the reconstruction speed. And the local similarity representation step make the method have better ability to suppress noise for applications under severe condition. As a general framework, other useful algorithms can also be incorporated into it conveniently. Experiments on commonly used face database demonstrate our scheme has better performance, especially for noise face image.

In this paper, we present a robust face recognition method with combined locality-sensitive sparsity and group sparsity constraint. The group sparsity constraint is designed to utilize the grouped structure information embedded in the training data. Its key idea is to try representing the test image with training images from fewer individuals. We show that, by further integrating the local similarity information between the test image and training images, the embedded group structure information can be better utilized, and as result, the recognition performance can be significantly improved. Experimental results on the ORL, AR and Extended Yale B database verify the superiority of our proposed method under different pose, illumination, expression variations and different dimension reduction settings.

Background prior has been widely used in many salient object detection models with promising results. These methods assume that the image boundary is all background. Then, color feature based methods are used to extract the salient object. However, such assumption may be inaccurate when the salient object is partially cropped by the image boundary. Besides, using only color feature is also insufficient. We present a novel salient object detection model based on background selection and multi-features. Firstly, we present a simple but effective method to pick out more reliable background seeds. Secondly, we utilize multi-features enhanced graph-based manifold ranking to get the saliency maps. Finally, we also present the salient object segmentation via computed saliency map. Qualitative and quantitative evaluation results on three widely used data sets demonstrate significant appeal and advantages of our technique compared with many state-of-the art models.

Lane marking detection is a basic task of Driver Assistance Systems (DAS) and Autonomous Land Vehicle (ALV). In order to improve the accuracy of lane marking detection, we design a rough lane marking locating method based on predecessors’ work. Considering the characteristic of lane markings, we extract Haar-like features of lane marking regions and train a strong cascade classifier by Adaboost Algorithm. The classifier is simple in principle and can fast locate the possible areas of lane markings accurately. Experimental results show that our method performs well.

Being as an important nonlinear component of block ciphers, Substitution box (S-box) directly affect the security of the cryptographic systems. It is important and difficult to design cryptographically strong S-box that simultaneously meet with multiple cryptographic criteria such as bijection, non-linearity, strict avalanche criterion (SAC), bits independence criterion (BIC), differential probability (DP) and linear probability (LP). To address the issue, an S-box generation approach based on L-L cascade Chaotic Map and Line Map (LLCMLM) is proposed in this paper. L-L cascade chaotic map is used to generate an integer sequence ranging 0–255, and line map is applied to scramble the position of the integer sequence. A series of experiments have been conducted to compare multiple cryptographic criteria of LLCMLM with other algorithms. Simulation results indicate that LLCMLM meets well with the design criteria of the S-box.

Scene text recognition is gaining renewed interest owing to the increase of scene image based applications and new intelligent devices. Unlike recognition of printed text, scene text recognition is challenging due to the complexity of scene images. To provide an overview of the techniques and inspire future research, this paper reviews the advances in scene character and text recognition, with emphasize on character feature representation methods and word recognition models. The papers published in the most recent conferences ECCV 2014, ACCV 2014, ICIP 2014, and ICPR 2014 are also reviewed in this paper to provide the state-of-the-art of scene character and text recognition. The state-of-the-art performance is provided to show the achieved performance so far and demonstrate the potential of deep learning based methods.

Spectral-Domain Optical Coherence Tomography (SD-OCT) is a non-invasive imaging modality, which provides retinal structures with unprecedented detail in 3D. In this paper, we propose an automated segmentation method to detect intra-retinal layers in SD-OCT images around optic nerve head acquired from a high resolution RTVue-100 SD-OCT (Optovue, Fremont, CA, USA). This method starts by removing all the OCT imaging artifacts including the speckle noise and enhancing the contrast between layers using the 3D nonlinear anisotropic. Afterwards, we combine the level set method, k-means and MRF method to segment three intra-retinal layers around optical nerve head. The segmentation results show that our method can effectively delineate the surfaces of the retinal tissues in the noisy 3D optic nerve head images. The signed and unsigned significant differences between the segmentation results and the ground truth over optic nerve head B-scans are 1.01

$$\pm $$

1.13 and 1.93

$$\pm $$

2.21.

This paper proposes a simple, accurate, and robust approach to single image blind super-resolution (SR). This task is formulated as a functional to be minimized with respect to both an intermediate super-resolved image and a

non

-

parametric

blur-kernel. The proposed method includes a convolution consistency constraint which uses a non-blind learning-based SR result to better guide the estimation process. Another key component is the bi-

ℓ

0

-

ℓ

2

-norm regularization placed on the super-resolved, sharp image and the blur-kernel, which is shown to be quite beneficial for accurate blur-kernel estimation. The numerical optimization is implemented by coupling the splitting augmented Lagrangian and the conjugate gradient. With the pre-estimated blur-kernel, the final SR image is reconstructed using a simple TV-based non-blind SR method. The new method is demonstrated to achieve better performance than Michaeli and Irani [

2

] in both terms of the kernel estimation accuracy and image SR quality.

Based on the Smoothed Particle Hydrodynamics (SPH) and Finite Element Method (FEM) model, we propose a method for real-time simulation of fluid with deformable bodies. The two-way coupling method for the fluid with deformable bodies is processed by the ray-traced collision detection method instead of the ghost particles. Using the forward ray-tracing method for both velocity and position, different normal and tangential conditions can be realized even for the cloth-like thin solids. The coupling forces are calculated based on the conservation of momentum and kinetic energy. In order to take full advantage of the computational power in modern GPUs, we implement our method in NVIDIA CUDA and OptiX. The simulation results are analyzed and discussed to show the efficiency of our method.

In this paper, we propose a novel model, combining the physical model and the visual model (visual-physical model), to describe the formation of a haze image. We describe the physical process of degraded image based on incorporation of optical imaging physical model and visual cognitive process, enriching the degradation factor. The variational approach is employed to eliminate the atmospheric light, then estimate transmission map via the median filter. We can recover the scene radiance based on MRF model, and use contrast limited adaptive histogram equalization to correct colors after defogging. Experimental results demonstrate that the proposed model can be applied efficiently to outdoor haze images.

In this paper, we propose an algorithm to improve some important details of sparse representation based image super resolution (SR) framework. Firstly, a new dictionary learning technique K-Eigen decomposition (K-EIG) is proposed. It improves the classical K-SVD algorithm in dictionary atom updating. K-EIG accelerates the learning process and keeps the similar performance of the learned dictionary. Secondly, image patch classification and edge patches extension are integrated into the SR framework. Two over-complete dictionary-pairs are trained based on K-EIG. In reconstruction, the input low resolution (LR) image is split into patches and each one is classified. The patch type decides which dictionary-pair is chosen. Then the sparse representation coefficient of the LR signal is inferred and the corresponding high resolution (HR) patch can be reconstructed. Experimental results prove that our algorithm can obtain competitive SR performance when compared with some classical methods. Besides, the time-consuming of dictionary-pair learning is lower.

Remote sensing images are extensively applied in various fields, however, they usually suffer from haze pollution, which always leads to low contrast and color distortion. In this paper, we propose a novel and effective spatial and spectral self-adaptive haze removal model for remote sensing images. Our research is based on the dark channel prior, meanwhile, we ameliorate the prior in two aspects as follow: first, to remove uneven haze in remote sensing images, we modify the original constant

$$ \upomega $$

to a matrix, and the value of

$$ \upomega $$

changes with density of haze, so the processing intensity varies as haze density changes. Second, the dark channel prior has severe color distortion when dealing with bright landforms, to solve this problem, we separate these bright areas automatically from other landforms and handle them alone. Experimental results reveal that our proposed model is able to remove uneven haze and keep good color consistency when dealing with bright landforms. Both visual effect and quantitative assessment demonstrate that the proposed algorithm is effective.

How to achieve the correct detection result for the infrared small targets is the important and challenging issue in infrared applications. In this paper, a small infrared target detection method based on low-rank representation is proposed, which used the low-rank representation (LRR) to decomposed infrared image to background component and target component, then the detection task could be finished through threshold processing. In different experimental conditions, the results show that our method based on low-rank representation not only has higher detection performance but also reduce the false alarm rate effectively.

This paper proposes a combined detector containing the background subtraction and the object appearance model-based detector. This is used to solve such problems as linking, overlapping, false object detecting etc. Then, we give a non-linear multi-mode tracker with the combined detector to solve such problems as sudden appearance changes and long-lasting occlusions, etc. Finally, we test our proposed person tracking framework in multi-object tracking scenario. Experimental results demonstrate that our proposed approaches have promising discriminative capability in comparison with other ones.

Liver segmentation in computed tomography (CT) images is a fundamental step for various computer-assisted clinical applications. However, automatic liver segmentation from CT images is still a challenging task. In this paper, we propose a novel non-parametric sparse representation-based deformation model (SRDM) for atlas-based liver segmentation framework using nonrigid registration based on free-form deformations (FFDs) model. Specifically, during atlas-based segmentation procedure, our proposed SRDM provides a regularization for the resulting deformation that maps the atlas to the space of the target image, constraining it to be a sparse linear combination of existing training deformations in a deformation repository. We evaluated our proposed method based on a set of 30 contrast-enhanced abdominal CT images, resulting in superior performance when compared to state-of-the-art atlas-based segmentation methods.

The image processing is becoming a key issue in astronomical data analysis. This paper introduces the algorithms and methods applied in astronomical image processing from different aspects. The paper first shows methods of lossless compression of astronomical images, like the pyramidal median transform, segment-based DPCM and 5/3 integer wavelet transform. Secondly it shows the algorithms of astronomical image segmentation, edge detection and de-noising. Finally, many different methods of image recovery and restoration are introduced briefly. We summarize a number of recent achievements on astronomical image processing in this survey, and list the recent published papers.

Nature light has a dynamic range beyond the modern camera can capture. So lots of commercial software applies techniques to reconstruct high dynamic range (HDR) image from low dynamic range (LDR) images. One of the popular methods is to recover the radiance of scene from LDR stacks [

2

]. However it needs to know the exact exposure time of LDR images and costs much time to solve the camera response function. These defects make it impossible to be built in hardware or real time systems. In this paper, we propose a new technique to synthesize HDR image based on the

intensity mapping function

(IMF). We first solve the IMF based on

cross

-

histogram map,

and then synthesize the pixel values from LDRs as if they have the same exposure time. We test our technique on several demos and get satisfying results with good visual quality in bright and dark scenes. Besides, because our method costs less time than the ‘software’ method, it is more suitable for further hardware application.

The reconstruction of the missing information of optical remote sensing images contaminated by unwanted cloud has attracted a great deal of attention. However, in practice, cloud removal is a challenging problem. In this paper, we propose to reconstruct the missing information by temporal domain group sparse representation. With the help of temporal normalization, the temporal complementation of multitemporal remote sensing images is strengthened. The group sparse representation, which seeks similar patches from the temporal domain, is then applied to recover the missing information. The experiments demonstrated that the proposed method is both quantitatively and qualitatively effective.

Head movement is an essential constituent of Chinese Sign Language (CSL), which helps to complete the definition of signing gestures and to assist in sending messages. Adding head motions into signing animations benefits for both the reality and the intelligibility. By analyzing the head motions both defined in words of CSL and captured from large motion data of a real signer performance, this paper proposes a quintuple for formalized head movement description. A Text To Head Motion (TTHM) synthesis model is established to perform a low-level semantic mapping from words to head gestures. Experimental results verify that improvement is achieved both in naturalness rating and understandability of signing animations after synthesizing with head motions.

A real-time machine vision detection system based on computer for fabric defect detection is presented in this paper. Hardware platform and software algorithm are the two main parts included in it. In hardware platform, image acquisition subsystem and transmission operated synchronously to achieve synchronization between motion and acquisition through the encoder and video capture card. Moreover, double-buffer technique with an alternative acquisition mode is applied to make the system more real-time. Each defect detection algorithm is regarded as a single detection unit which is integrated in the software system. Then different detection units are employed at different fabrics and defects to gain better detection efficacy. It could be concluded that the proposed system provides a lower cost, higher performance and more excellent expansibility solution for enterprises via the variety of experiments.

In the video surveillance system, due to the complicated background, the targets in the movement process often appear some or full of occlusion. How to detect occlusions, handle of issues efficiently, especially in the event of long time occlusion. Accurate target identification and tracking is the key indicators to evaluate the robustness of a target tracking algorithm. This paper deals with long time occlusions based on the concept of “object permanence” in psychology. This paper proposes a method based on “object permanence” algorithm to solve the identification and tracking problems after long time occlusions. The experimental results show that this algorithm can effectively solve the occlusion problem.

We present a fully automated method to segment the interventricular septum from cardiac MR images in this paper. By introducing the circular Hough transformation our model can automatically detect the contours of left ventricle as circles used as the initialization. The interior and exterior energies are weighted by the entropy, which improves the robust of the evolving curve. Local neighborhood information is used to evolve the level set function, which can reduce the impact of the heterogeneous grays inside of regions and improve the segmentation accuracy. The adaptive window size is utilized to reduce the sensitivity to initialization rather than a fixed window size. The Gaussian kernel is used to not only ensure the smoothness and stability of the level set function, but also eliminate the traditional Euclidean length term and re-initialization. Finally, we segment the septum automatically by the classical segmentation methods combined with anatomical location information. Extensive experiments indicate that the superior performance of the proposed method over the state-of-the-art methods in terms of both good robustness and high efficiency.

Convolutional Neural Networks (CNN) are an alternative type of deep neural network that can be used to model local correlations and reduce translation variations, which have demonstrated great performance in some computer vision areas except the visual tracking due to the lack of training data. In this paper, we explore applying a two-stage learning CNN as a generic feature extractor offline pretrained with a large auxiliary dataset and then transfer its rich feature hierarchies to the robust visual tracking task. Instead of traditional neuron models in CNNs, we introduce a strategy to use ReLU for training acceleration. Empirical comparisons prove our CNN based tracker outperforms several state-of-the-art methods on an open tracking benchmark.

Dynamic magnetic resonance imaging (dMRI) is widely used in human motion organ and functional imaging. But it requires reducing the imaging time to obtain high spatial and temporal resolution. This paper proposes a double sparse spatiotemporal dictionary model for compressed sensing reconstruction of dMRI from undersampled data. The model extends the ordinary 2-D dictionary to 3-D spatiotemporal dictionary by sparse representation of both the signals and dictionary atoms. Specifically, the first level sparse representation of dictionary atoms is learned with K-SVD algorithm. The second level sparse representation of spatiotemporal patches is obtained by OMP algorithm. An alternate iterative optimization is applied to solve the problem. Experiment results demonstrate that comparing with the state of the art method k-t FOCUSS and single level dictionary learning – DLMRI, the proposed method performs better in removing aliasing artifacts and in capturing temporal variations as well.

This paper describes a novel method to recover underwater images by devignetting and colour correction. Scattering and colour distortion are two major problems of degradation for underwater imaging. Scattering is caused by large suspended particles. Colour distortion corresponds to the varying degrees of attenuation encountered by light traveling in water with different wavelengths, rendering ambient underwater environments dominated by a bluish tone. To this end, we propose a novel underwater imaging model, which is much closer to the light propagation model in underwater environment. We remove the noise by dual-tree complex wavelet transform. Then, solve the non-uniform illumination of artificial lights by devignetting. Finally, we recover the image colour through camera spectral responses. The corrected images are characterized by reduced noised level, better exposedness of the dark regions, improved global contrast while the finest details and edges preserving.

We propose a system for video based face tracking and animation. With a single video camera, our system can accurately track the facial feature points of a user, and transfer the tracked facial motions to the avatar’s face. We use constrained local model (CLM) to track the feature points. The original CLM only makes use of local texture and performs an exhaustive local search around the current estimate of feature points. This often leads to local minima. To overcome this problem, we incorporate the global texture into CLM. The improved CLM not only gives discriminative capability to each feature point, but also gives good match to the whole texture. After obtaining the 2D positions of the feature points, we estimate blendshape coefficients based on a set of user-specific 3D key shapes. Finally, facial animations are created using blendshape interpolation. Experiments demonstrate the effectiveness of our system.

Video stabilization has been gaining in importance in earth observation and video surveillance on boarded of airship platforms. As the airship platforms have obvious low-frequency vibration caused by wind and the unstable attitude may also lead to geometric distortion in the video streams, and most of the traditional methods are designed for casual handheld devices which cannot effectively handle those kinds of problems, a new video stabilization method for airship earth observation is proposed. In this method, a modified camera path planning method based on piecewise linear L1 optimization is given under the specific motion properties of airship platform and the needs of geometric correction for each frame, and the camera path reconstruction is carried out with vision based motion estimation and sparse GPS and attitude data. The effectiveness of our method is confirmed by quantitative experiments over a variety of video streams.

Ever increasing populations are putting considerable strain on the critical infrastructures of our towns, cities, and countries. The interconnecting and interdependent components of these man-made living procedures and protocols give-way in unforeseen, unplanned situations. Having the ability to visualise these interconnecting entities and the interaction they have on one another is critical for future city planners. We propose a novel framework called Project Vision Support that provides an automated visualisation of real world open data maps for the creation of procedurally generated urban environments. This framework can then be used to implement planning and scheduling algorithms for the orchestrated task of emergency services for crisis management response.

One of the most important problems in image editing applications is how to preserve the important structure of image.Local linear model is widely applied to image editing application such as image filter, alpha matting. It preserves the local linear structure in image which describes the local feature of image.

In this paper, we propose the nonlocal nonlinear model in image editing method. In the nonlocal area of image, nonlinear structure is extracted from images by least square method and kernel trick.Different from local linear model, nonlocal nonlinear model can represent the nonlinear structure in nonlocal area of image. It can be widely applied to image denoising, image upsampling, alpha matting. Results show our model is effective.

Crisis management is a modern phenomenon brought about by natural disasters and acts of terrorism. Building a modern crisis management response program needs a multi-disciplinary architecture and accurate, up-to-date, real-world data. The creation of virtual environments depicting critical infrastructure buildings and conduits between these highly interconnected man-made structures is a complex procedure. The crossover between games technology and use of real-world map data for real-world simulations is becoming more common with the advancements of computer hardware and software, and the accuracy of real-world map data. However, there are many problems with using real-world map data for simulation due to the large potential of missing and error prone data involved in this big data. Within this work we use three types of data sets; Ordnance Survey data, LiDAR data, and OpenStreetMap data to provide accurate map and 3D environment information for crisis management systems. Combining these large data-sets can reduce errors and retrieve missing data for use within a modern game engine for visualization analysis. We propose a novel technique for data extraction using adjacency matrices for custom model generation corresponding to real-world structures such as landscapes, buildings, road systems, area boundaries, or a combination of these at different resolutions.

The research of image memorability has received increasing attention recently. In this paper, the influence of the visual attention based features on image memorability is explored, which is different from most of the existing studies focusing on various appearance features. In this paper, the dataset used by Isola et al. are adopted. The visual saliency map of each image in the dataset is generated via the visual attention model. The corresponding object-saliency map is obtained by replacing each object with its average visual saliency. The global, local, and joint spatial histograms based on the object-saliency map are obtained and the relationship between visual attention and memorability is explored based on these visual attention based features. The experiments are carried out by using two existing visual attention models and demonstrate that these mentioned visual attention based features are more effective than the appearance features to predict the image memorability.

On a daily basis, urban residents are unconsciously exposed to hazardous noise levels. This has a detrimental effect on the ear-drum, with symptoms often not apparent till later in life. The impact of harmful noises levels has a damaging impact on wellbeing. It is estimated that 10 million people suffer from damaged hearing in the UK alone, with 6.4 million of retirement age or above. With this number expected to increase significantly by 2031, the demand and cost for healthcare providers is expected to intensify. Tinnitus affects about 10 percent of the UK population, with the condition ranging from mild to severe. The effects can have psychological impact on the patient. Often communication becomes difficult, and the sufferer may also be unable to use a hearing aid due to buzzing, ringing or monotonous sounds in the ear. Action on Hearing Loss states that sufferers of hearing related illnesses are more likely to withdraw from social activities. Tinnitus sufferers are known to avoid noisy environments and busy urban areas, as exposure to excessive noise levels exacerbates the symptoms. In this paper, an approach for evaluating and predicting urban noise levels is put forward. The system performs a data classification process to identify and predict harmful noise areas at diverse periods. The goal is to provide Tinnitus sufferers with a real-time tool, which can be used as a guide to find quieter routes to work; identify harmful areas to avoid or predict when noise levels on certain roads will be dangerous to the ear-drum. Our system also performs a visualisation function, which overlays real-time noise levels onto an interactive 3D map.

We present a novel fitness training system called Heart-Creates-Worlds (HCW) as a practical while pervasive solution to encourage effective daily physical activity. In this system, the aesthetic audiovisual effects of a virtual world are tightly associated with the user’s real time heart rate, while the heart rate reflects the users’ physiological and psychological state accordingly. By physically act on the control of the representation of the virtual world, users are being persuaded into traveling in a more aesthetically pleasing virtual surroundings than in the displeasure one, meanwhile, naturally fix themselves at the target fitness training zone. A pilot user study was conduct to evaluate the effectiveness and enjoyment of HCW. The results indicate that, this kind of aesthetic-driven fitness training system is helpful on encouraging regular physical performance with enough amounts.