Advances in Image and Graphics Technologies

Automatic, fast and accurate extraction of the blood vessel is an important task in the image-aided diagnosis of disease. In this paper, we describe a novel superficial vessel imaging and projecting system. First, the superficial vessel of human arm is captured by NIR imaging technique, and then three pre-processing algorithms, including hair removal, non-uniform illumination correction and vessel enhancement, are developed to strengthen the vessel image. Second, a model-based binarization method is proposed to detect vessel-like structures. And then the mathematic morphological and connected component refinement methods are integrated to remove small vessel noises. Experimental results demonstrate that our system can imaging superficial vessels on real-time. The proposed system does not need any human interaction, which hence can be used in clinical venipuncture practices.

Looking for the best solution for Confrontation is crucial for the effect of electronic countermeasures. In this paper, we put forward the concept about the optimal laying space [1], analyze and evaluate the simulation results in different countering situation base on the maritime infrared electronic countermeasures simulation platform [2]. Then, here comes up with a way to get the optimal laying space based on PCA [3], which effectively solved the problem of strategy optimazation in electronic countermeasures. Also, the paper builds a model to evaluate the probability of countering success by using the multiple regression analysis [4], and provides the probability parameter for the selected best countering strategy. The testing results match properly with the results from the simulation platform, indicating that this method is able to find the best strategy for electronic countermeasures.

With the development of digital media technology, the problem of digital product copyright protection is becoming more and more serious. The second generation of watermarking technology makes outstanding contributions to solve this problem. This article studies image edge histogram descriptor algorithm to analyze the image feature, then uses JND model which analyses characteristics of human perception to conduct watermarking embedding. The experiment results show that our watermarking algorithm has strong robustness and an acknowledgment.

Visual target tracking is one of the key technologies to implement full automatic exploration for a planetary rover and improve exploration efficiency. A novel visual tracking system is developed based on the Tracking-Learning-Detection (TLD) algorithm in combination with stereo image matching to achieve 3D tracking of a science target. Experimental results using stereo image sequences demonstrate the excellent performance of TLD tracking and the overall effectiveness of the 3D tracking.

In this paper, real-time water rendering approaches using the graphics hardware is described. The key optical and motion characteristics of water are present. Although, the complex optical behavior and physical interactions can be calculated absolutely accurate, the computational capacity of today’s graphic cards is limited; the optimal compromise between realism and accuracy can be different depending on the target platform and required result. Perlin noise, which is one of the most commonly used approaches to simulate the wave animation, is introduced to avoid unnecessary computations. Rendering reflections and refractions is generally achieved by cube-maps simply. The demo applications demonstrate some discussed approaches and can be adjusted to various expectations.

In the application of face recognition, with the increasing number of stored face mode in ART2 network, it will spend a lot of time to learn or identify the future entering mode of ART2 network, and then the speed of face recognition will become slower. The author proposed an improved ART2 algorithm based on rough classification, using the adaboost algorithm to train a classifier to determine whether the face wearing glasses, the face mode will be divided into two categories of people who wear glasses and do not wear glasses by deciding a people whether to wear glasses. The experiments show that the method can greatly improve the speed of face recognition.

The performance of motion estimation is of great importance for H.264 advanced video coding. It is estimated that motion estimation consumes about 70% of the encoding time. Lots of motion estimation algorithms are proposed to improve the encoding speed. Unlike the assumption of most motion estimation algorithms, the horizontal motion vectors are much larger than the vertical in most cases. With the unsymmetrical characteristic, this paper presents a new diamond search based motion estimation algorithm to improve the efficiency. The unsymmetrical diamond search shorten the vertical step to lower down the computation complexity. The search points of the big template and the small template are reduced to 5 and 3 respectively. The simulation results show that, the unsymmetrical diamond search can achieve much more significant speedup ratio than other motion estimation algorithms with relatively high probability.

This paper introduces a new method for fusing palmprint and palmvein. The focus of image fusion is placed on the application of neurodynamics of vision models, and the characteristics of the bimodal cells of rattlesnakes are researched. An image acquisition device which can collect palmprint and palmvein images in the same position and at the same time is deviced, and a small palmprint and palmvein database is built by this device. Some experiments have been done to evaluate this fusion method on the palmprint and palmvein database.

Due to Chinese traffic safety situation which is becoming more and more serious, this paper presents a fatigue detection method based on machine vision to provide safety information for the driver in the process of driving. It uses difference, gray projection and the complexity and real-time image processing techniques to detect and analyze the state of eyes, calculate the blink frequency to determine whether the driver is fatigue or not. The results of the experiment show that the method has higher detection accuracy.

A new method about the extraction of cirrus cloud and its shadow from multispectral imageries was proposed in this paper, which based on the reflection characteristics, shape and position relationships of cirrus cloud and its shadow. First, analyzed their reflection characteristics, and then used two thresholds to attain the general regions. Second, based on their shape and position relationships, used the dilation operator and image shift method in turn to attain their accurate regions. The new method was tested with the imagery of HJ-1A/B in Fuling Chongqing, and compared with the traditional methods was the best.

This paper presents a novel patch-based approach to still image denoising by principal component analysis (PCA) with geometric structure clustering. Inspired by denoising image patch-wise ideas, we decompose it to overlap patches which contain different content and structure information. However, some of them have similar geometric dominant orientation. In order to cluster the geometric patches, we utilize their gradient map to compute the dominant orientation of the gradient field. Such a clustering procedure guarantees that only geometric patches with similar dominant orientation are used to perform hard thresholding on the coefficients in the PCA domain to remove the noise. We carry out a comprehensive empirical evaluation of the performance of this algorithm in terms of accuracy and visuality. The results reveal that our method appears to be competitive with the state-of-the-art denoising algorithms.

In this paper, based on multi-temporal remote monitoring technology, using object-oriented classification method to monitor the change of vegetation of Zhangye oasis from TM/ETM data in 1989, 2000, 2011 years. The results show that: (1) the multi-resolution segmentation converted the single cell which had the similar texture, spectrum and shape to the object. Integrating nearest neighbor classifier and membership classifier to class the three data, and the overall accuracy of classification was 89.5%, Kappa coefficient was 0.9. The classification stability was 0.45 and 0.47 in 2000 and 2011 years. It showed that object-oriented classification method accuracy is higher than traditional classification method. (2)The three classification results indicate the area of bare land was larger than other, and it was reducing, with a percentage was 73.21%, 64.76%, 60.17%. The vegetation mainly distributed in both sides of Heihe River, the percentage of three data were 16.01%, 29.9%, 33.6%. The saline land was mainly distributed in the northwest of the oasis region, the percentage dropped to 2.33% from 4.89% during 1989 and 2011 years. (3) NDVI of the upstream was higher than the NDVI of the downstream on sides of river, the NDVI raised and the maximum value was 0.54. NDVI increased significantly from 1989 to 2011 years in Linze central region, and the maximum value reached to 0.58 in 2011 years, and it had the same characteristic in the southeast of Ganzhou district. The average NDVI of 2011 years was higher than in 2000 and 1989.

In this paper, we introduce a new detection method for extremely weak moving target in infrared image sequences based on the novel three-dimensional directional filtering. The main points of the method are, first, we use a dual-diffusion partial differential equation (DFPDE) to pre-whitening an image, which can suppress the constructive texture background effectively and keep the target signal steadily. And second, to match precise target motion characteristic, we propose a Wide-to-Exact search method that can improve the speed of filtering. Experiment results demonstrate that our method can perform good detection results, even at poor signal-to-noise ratio.

In this paper, a new blind video watermarking scheme in H.264 compressed domain is proposed based on the existing video watermarking schemes and combing with the new characteristics of H.264 coding standard. The submacroblocks of I4 which have more nonzero quantized coefficients are chosen as embedding locations. The watermark is embedded into video sequence by modifying the number of nonzero (NNZ) quantized coefficients of I frames on the basis of the probability matrix which is defined according to the distribution characteristics of DCT coefficients. The simulation results show that the proposed method can prevent bit-rate increase and improve embedding capacity without sacrificing perceptual quality. On the other hand, it is not necessary to fully decode the compressed video in extracting process, and this method can also meet the requirements of real-timing and blind detection.

Virtual Reality (VR) technology is a highly comprehensive architectural design technique of expression. With the development of building information and modeling of digital technology, VR technology provides a broad platform in the application of restoration of cultural architecture. Many cultural relic buildings have disappeared or have been dilapidated during Sichuan Earthquake. In order to enable people to really appreciate the styles of ancient buildings and their profound historical and cultural connotation, we apply VR technology to reproduce the ancient buildings and show the artistic and cultural messages what the ancient buildings convey by recovering, simulating, reproducing, displaying and preserving the ancient buildings through computer.

In the cloud storage service, according to the data on the cloud computing safety protection problem, the paper presents secure obfuscating homomorphism encryption scheme. Constructing a point function obfuscation that based on perfectly one way probability hash function in scheme, construction depends on hash function and the computational difficulty problems, then use the computational difficulty problems, to realize the encrypted homomorphism function, also guarantee the function of the point function obfuscator at the same time, the scheme raises the security of the encrypted data. This paper provides the security proof of the scheme, shows that the scheme is feasible.

A simulative tank shooting method based on image analysis is presented, it is an alternative to the existing laser system. The system is composed of an image acquisition system and an infrared strobe light system. Based on the strobe light, the ID of the target tank can be computed. Analyzing the characteristics of the target image, the position of strobe light is determined through image normalization, image segmentation and tracing boundary contours algorithms in turn; then the target ID is identified. Experimental results show that the proposed method can recognize the ID and hit location of the target tank accurately.

In this paper, we propose a novel Scale Noise Level Estimation method based on Double-Density Dual Tree Complex Wavelet Transform (DD_DTCWT), which is referred to as DD_DTCWT_SNLE, to take the advantage of the correlation between the noise and noisy coefficients of DD_DTCWT. The novel DD_DTCWT_SNLE method is formulated through both theoretical analysis and numerical simulation, and is applied into three different threshold de-noising schemes respectively. Simulation results show that there is an approximate linear relation between DD_DTCWT_SNLE and the noise level and that DD_DTCWT_SNLE can reflect the noise level of coefficients in each layer more accurately. The proposed method outperforms the bivariate shrinkage algorithm and a gain of 0.8 dB in PSNR is obtained when compared to other DD_DTCWT based algorithms. We also show the universal applicability of our DD_DTCWT_SNLE for multi-scale linear operators, and its usage as a noise level estimator for all the other linear multi-scale decomposition coefficients.

Large amounts of complex data will be produced when detecting pesticide residues in edible agricultural products within some areas, such as in our whole county. .In order to helping experts and decision-makers understand and analyze these large amounts of data accurately and effectively, a treemap-based visualization method for pesticide residues detection data is presented in this paper. First, the characteristics of pesticide-residue detection data are analyzed. Then, a treemap-based visualization method is presented. It uses hierarchical data visualization technique based on treemap combining with interactive techniques, such as detail viewing, selecting and filtering. The results of applying this method to analyzing pesticide-residue detection data set demonstrate that it can help experts to analyze data set according to the hierarchical structure of regions and categories of agricultural products effectively.

Expression recognition is an important subjective measurement method in emotional calculation. Considering the stability and representativeness of image algebra features, expression characteristics can be extracted by singular value decomposition. BP neural network optimized by the genetic algorithm is adopted as a classifier. Using the classifier, an expression recognition experiment was done on the JAFFE library and emotional induced experimental expression database. Comparing with the traditional BP classifier, the results of the experiment proves that the method is more effective and efficient.

The block adaptive quantization(BAQ) algorithm is comparatively mature for SAR raw data compression at present. This algorithm is on the premise that SAR raw data should satisfy Gauss distribution. But the imaged region is quite rugged, some blocks of data doesn’t satisfy Gaussian distribution. Therefore, a block adative scalar-vector quantization(BASVQ) algorithm is put forward in this paper, namely, scalar quantization is applied when data blocks satisfy Gaussian distribution while vector quantization is applied when don’t satisfy. The experiments demonstrate that the performance of BASVQ algorithm outperforms that of BAQ algorithm. The BASVQ algorithm has practical value in some degree.

Moving objects detection is a fundamental step for automated video analysis, robot visual system and many other vision applications. There are limitations in the existing algorithms, such as assuming a static camera, a smooth motion and rigid motion of target objects, etc. In this paper, we present a novel model named IRTSW-model; a moving objects detection model which can work effectively no matter the camera is moving or static. In the approach, images registration is used to eliminate the relative movements between the background and the camera; unsupervised codebook model is constructed to model the background; and then the moving objects are detected accurately. Experiments on the segtrack database demonstrate the effectiveness of our model.

Due to providing a new solution for the determining the overall status of plants, computer vision in plat monitoring is attracting more and more attention. This paper reviews recent advances of computer vision in the detection of plant stresses. Firstly, this paper reviewed image segmentation for separating the candidate’s green plant material. Then, the detection approaches were summarized from the aspects of water stress, nutrient stress, diseases and pets stress. Emphasis is placed on the way of image analysis and feature extraction. The merits and drawbacks of the approaches were discussed too. The computer vision system, which could be adaptive to the varying of natural light and could detect different feature of the plant’s stress in complicated background, would be promising and valuable. The difficulties and research interesting of computer vision in plant stress detection area in future were given in the conclusion.

This paper provides a framework that defines a set of mapping symbols for representing the emergency events and related operated behaviors. In order to support decision-making effectively through emergency response, a framework was designed by involving two key elements, visual display and interactivity, to simulate the decision-making process. Visual display element is used to describe the crisis event under the background of spatial information environment, while interactivity element stands for the operations applied in the visual spatial layer. To evaluate the feasibility, a prototype system for marine disaster management is developed.

In the object recognition process of infrared image, as the amount of training data is very small, traditional learning does not construct a high-quality classifier for the recognition object. Aimed at the problem, a transfer knowledge framework for object recognition of infrared image is proposed in this paper. Hu moments is firstly extracts as feature vectors of object data, and then a large amount of exist object data with different distributions to the recognition object data is seen as the auxiliary training data in the feature spaces. Our transfer knowledge approach can transfer knowledge from the auxiliary data to help the tiny amount of training data to train a better classifier, which improve the performance of object recognition. According to the experiments in infrared images, it shows that the accuracy of object recognition has been greatly improved by our proposed approach compared with the other classical methods.

Localization of the rover and mapping of the surrounding terrain with high precision is critical to surface operations in planetary rover missions, such as rover traverse planning, hazard avoidance, and target approaching. It is also desirable for a future planetary rover to have real-time self-localization and mapping capabilities so that it can traverse longer distance and acquire more science data. In this research, we have developed a real-time high-precision method for planetary rover localization and topographic mapping. High precision localization is achieved through a new visual odometry (VO) algorithm based on bundle adjustment of an image network with adaptive selection of geometric key frames (GKFs). Local topographic mapping products are generated simultaneously in real time based on the localization results. Continuous topographic products of the entire traverse area are generated offline. Field experimental results demonstrate the effectiveness and high-precision of the proposed method.

The Nearest-Neighbor (NN) rule is a widely used technique for pen gesture recognition. It is quite simple and effective, furthermore, it allow for flexible extension of newly defined classes and training samples. However, the expensive cost of computation is a key issue need to be addressed. A fast template matching method for pen gesture recognition was proposed. We embed gesture description vectors nonlinearly into a low-dimensional space. The square Euclidean distance in the embedded space can be used as the lower bound of the distance in the original space, thus most dissimilar training templates could be rejected in the embedded space. Experiment results show that our method is almost over an order of magnitude faster than the naïve NN while achieving comparable recognition accuracy. Moreover, it does not rely on complex data structures or preprocessing, making it more suitable for applications that call for dynamic training data, such as user adaptive recognition.

In this paper, giving an example for HuaShan rock painting in GuangXi Zhuang Autonomous Region, a new method for nondestructive detection of stalactite disease has been tried out by means of infrared thermal imaging technology. Based on heat conduction theory, the finite element numerical simulation of the stalactite diseases are done by finite element software ANSIS11.0. The simulation results validate the feasibility of infrared thermal imaging technology for nondestructive detection of the rock painting stalactite diseases.

In this paper, a practical and efficient algorithm based on the triangulated polyhedra is proposed to calculate EV (Edge & Vertex) and EEE (Triple Edge) events for 3D viewpoint space partition. At first step a few triangular faces that contain distinct silhouette vertexes and edges are chosen in order to simplify the model, and then some of the EV and EEE events occluded by other faces is pruned by using the view-independent pruning algorithm. After the first step, the rest of EV and EEE events are actual critical events which are then calculated for space partition. Therefore we avoid calculating many EV and EEE events which are not actually existent before space partition so that it reduces computational complexity enormously. In the last section of this paper, we apply this method to two kinds of aircraft models and one kind of car model for experiments. The results show that it can effectively carry out calculation of EV and EEE events and space partition. And the representative viewpoints are placed over the viewpoint space evenly. On this foundation, actual 3D object recognitions could also be implemented.

Multi-light source optimization techniques is a technique to improve the rendering time for many-light source. Lightcuts[1] is a classic algorithm and can handle complicated scenes and get very good result with faster speed. In this paper, we made an improvement to the algorithm for the artifacts of soft shadows. In our method, we select more than one representative of lights in light clusters and choose one light as the representative randomly during the rendering process. The experimental results show that this method can effectively improve the quality of soft shadows.

This paper presents a variational framework for multi-region image segmentation method based on image structure tensor. The multi-region segmentation is addressed by employing the multiphase level set functions with constraint. The image feature is extracted by using the image structure tensor. The coupled Partial Differential Equations (PDE) related to the minimization of the functional are considered through a dynamical scheme. A modified region competition factor is adopted to speed up the cure evolution functions, it also guarantees no vacuum and non-overlapping between the neighbor regions. Several experiments are conducted on both synthetic images and natural image. The results illustrate that the proposed multi-region segmentation method is fast and less sensitive to the initializations.

It has been well accepted that compression is essential to the real-time visualization of large terrain dataset. Many efficient compression algorithms have been presented to enhance the performance of rendering. However, the issue of how to render the terrain after decompression is not well addressed. Most previous rendering methods are not well suitable when compression is involved. In this paper, to overcome this problem, we explore an alternative terrain rendering method which utilizes GPU tessellation. By moving the level of detail (LOD) unit to finer granularity, simplification of terrain models can be performed on regular grids, which is simpler to implement than irregular/semi-regular grids. Only a few indices are needed for the simplification of a dataset, thus will not negate the effect of compression. Experiments show that the method can effectively render large terrains with an error tolerance of one pixel.

The phenomenon that objectionable contents spread over the Mobile Internet reflects badly both on users and business. To cope with the situation here, we have proposed a relatively effective and efficient method. Combined with the conventional skin color detection and face detection, we add movement invariants to revise the detection ability and use image clustering based on MPEG-7 to improve the efficiency of human examination and verification. Simulations have shown the good performance for the realtime detection effects, and reduced the misstatement Rate and 90% artificial workload, which improve the detection ratio to a large scale.

In this paper, we use a piece of A4 paper as the modal plane, which not only reduces the requirement of the modal plane and makes the calibration very convenient, but also solves the point matching problem. Extended experiments have proved that this camera calibration method is available and effective.

In order to simulate the angular velocity, acceleration and other kinematic parameters during the training timely for the ski athletes, we design a three-dimensional motion simulation system based on the gesture information acquisition system, which can help team coach to analyze the action and guide the team members training with the given purpose. Since it is difficult to get the accurate position parameter of the ski athlete, we integrate the ski field and physical information. The gesture information acquisition system can gain the parameter information of three-dimensional acceleration and angular velocity, which are recorded by the sensor. In this paper, we construct a three-dimensional motion simulation system to achieve the parameters of velocity, angular acceleration and angle at any time. The three-dimensional motion simulation system for freestyle skiing aerials is constructed in the OpenGL environment, and can observe the effects of posture demonstration from different views. Simulation results show that this system can record the movement information accurately and help athlete obtaining the training targets.

Mammography is the most effective method for the early detection and diagnosis of breast cancer diseases. However, mammographic images contaminated by noise generally need image enhancement techniques to aid interpretation. The paper proposes a new image contrast enhancement method for digital mammograms based on guided image filtering method. Guided image filtering is a non-iterative, non-linear filter, which not only smoothes low gradient regions, but also preserve strong edges. Analogously to the bilateral filter, this filter has edge-preserving properties, but can be implemented in a very fast way. Thus, we adopt guided image filtering algorithm to increase the contrast in mammograms for human easily extracting of suspicious regions. Experimental results show that the proposed method gives superior image quality compared to other enhancement methods.

To improve the spatial resolution of image, many methods such as example-based and interpolation are presented nowadays. However, the smoothness of edges and the preserving of texture details in the zoomed image are still need to be improved to obtain better performance. Based on the residual correction and compensation idea, we propose a novel algorithm on image zooming, which refines the reconstructed residual using the Steer Kernel Regression and nonlocal filtering. Experimental results show that the proposed algorithm can not only well preserve the texture details but also enables the reconstructed image to achieve better visual effect and resolution.

Gray World algorithm is a classical method of Color Constancy, which is based on the assumption that the average reflectance of surfaces in the world is achromatic, but it usually fails when the image has a large dominant color patch. In this paper, we propose an improved algorithm based on Salient Detection to achieve Color Constancy, which can cover the aforementioned shortage of the Gray World algorithm. The extensive validation of our method on commonly used dataset which includes images under varying illumination conditions is presented. Experimental results demonstrated that our method is robust to the choice of dataset and at least as good as current state-of-the-art Color Constancy approaches.

Road traffic signs contain important traffic information so that the automatic location of traffic signs in the intelligent transportation system is especially important. This paper proposes an improved tracking method for automatic location of traffic signs in the image-based virtual space. Firstly, the regions of candidate traffic signs are extracted. Then moment invariants are used for traffic sign recognition. In Mean-Shift algorithm, the target model is initialized using the above recognition result, and Mean-Shift algorithm is used for tracking. On the basis of Mean-Shift algorithm, similarity of moment invariants between candidate model and standard templates are calculated for determining whether to update initial target model. Eventually, traffic signs are located through coordinate system transformation. The experiments show that the proposed method can not only effectively adjust the size of tracking window, update the target model, and improve the tracking result, but also satisfy the real-time requirement of our location system.

For WMS-based projection plane data this paper puts forward a comprehensive construct method and optimization technology of global 3D modeling by Quad-Tree Model. In that model, tile region is set by its layer index and grid index according WMS data organization characteristic. Followed, based on statistical methods, by setting constant value to coordinate layers index we implement layers optimal matching and keep layers of images tiles, DEM blocks and model nodes consistent with each other. Even more important, for the particularity of WMS-based projection plane data, we analyze tiles distribution characteristics in zonal direction, and find out a technical means to optimize tile mesh modeling based on area change curve segmentation dividing. At last, according to experiment and experience, we propose out dynamic segmentation strategy to optimize mesh size by fixed segments proportion array list and work out its implementation in detail. In order to verify the validity of this approach, this paper also gives the method of calculating the theoretical efficiency lifting and the assessed result value. In fact, in our specific application, efficiency improvements have been far beyond the 5 times since it integrates other optimization methods.