Table of Contents

30 June 2019, Volume 40 Issue 3 Previous Issue    Next Issue

---

For Selected:

Download Citations EndNote Ris BibTeX

Toggle Thumbnails

Select

False Positive Reduction of Pulmonary Nodules Using 3D CNN YOU Kun1,3, HAO Peng-yi1,3, WU Fu-li1,3, ZHANG Fan1, WU Jian2,3 2019, 40(3): 423-428.  DOI: 10.11996/JG.j.2095-302X.2019030423 Abstract ( 174 )   PDF (834KB) ( 171 )   Pulmonary Nodule Detection is the most promising way in early detection of pulmonary cancer. False positive reduction is one of the most crucial steps for improving the accuracy in automatic pulmonary nodule detection. For quickly and accurately discriminate true nodules from a large number of candidates, a 3D convolutional neural networks (CNN) is proposed for false positive reduction. In the proposed network, identity mapping and residual unit are adopted to accelerate network training. At the same time, single connected path is explored to get new features. Compared with 2D CNN based pulmonary nodule detection methods, the proposed method based on the proposed 3D CNN can take full advantage of space structure of CT. Compared with other 3D CNN based pulmonary nodule detection methods, the proposed method has fewer parameters that can make the training very fast. Experimental results on the public LUNA16 dataset demonstrate superior performance of the proposed method. Related Articles | Metrics

Select

Robust Uniform B-Spline Models for Interpolating Interval Data YANG Jing1,2, HAN Xu-li1 2019, 40(3): 429-434.  DOI: 10.11996/JG.j.2095-302X.2019030429 Abstract ( 81 )   PDF (487KB) ( 117 )   We adopted the uniform B-spline in building the robust models for interpolation problems of interval data. Compared with the robust models of the traditional polynomial spline, the B-spline models proposed in this paper have a number of advantages that enable a solution of an obvious finite convex optimization problem due to more concise formulations and a more simplified computing process. We put forward the method of creating the characteristic polygon vertices based on the given interval data for the purpose of increasing the freedom degree of models. The robust optimization models of the frequently used uniform quadratic and cubic B-spline are formulated in detail as finite convex optimization problems, which are applicable to the existing optimization algorithms and software. Some numerical examples are presented to show the feasibility and effectiveness of the proposed robust models. Related Articles | Metrics

Select

Structure-Based Real-Time Image Stippling LIU You-quan1, ZHANG Cai-rong1, MA Lei2,3, SHI Jian4, SUN Zhao1, CHEN Yan-yun3 2019, 40(3): 435-440.  DOI: 10.11996/JG.j.2095-302X.2019030435 Abstract ( 123 )   PDF (5397KB) ( 157 )   Stippling is a classical art technique that uses small dots to simulate varying degrees of solidity or shading and show its artistry. Generating stippling from image is one of the targets of image stylization and non-photorealistic rendering. In this paper, we propose a structure-based real-time image stippling algorithm. A pre-computed incremental Voronoi sequence is employed to generate sample points for stippling. To improve the quality on the edge and detail regions, we introduced input image and its corresponding image structure information to guide the radius and tone of the stipples. Experiments show that the stippling generated by pro-posed method is superior to the state-of-the-arts in visual and numerical error (SSIM). Our proposed algorithm can achieve real-time performance and apply to high frame rate scenes. Related Articles | Metrics

Select

Research on Gait Recognition Algorithm Based on Double Features Using the Layer Matching Fusion Method LI Hong-an1, DU Zhuo-ming2, LI Zhan-li1, HUI Qiao-juan3, BAI Jia-hao1 2019, 40(3): 441-446.  DOI: 10.11996/JG.j.2095-302X.2019030441 Abstract ( 66 )   PDF (495KB) ( 115 )   The gait recognition method is a kind of identity recognition method according to the walking postures in the distance. For the low recognition rate caused by the incomplete outline image and the selected key frame which would easily lose information, we propose a gait recognition method based on double features using the layer matching fusion method. The gaits have both the static image characters and the dynamic speed characters, and we use the layer matching method to fuse 6 invariable moment features of the Hu moment with the frame difference percentage features. Firstly, the Hu moment features and the frame difference percentage features are extracted from a period of the normalized gait images, and the 6 invariable moment features are described as one feature vector. Secondly, the layer matching fusion method is used to fuse the tow features. Lastly, the k-Nearest neighbor method is used for the identity recognition. The experiments show that our method could efficiently raise the recognition rate. Related Articles | Metrics

Select

Surface Segmentation Algorithm Based on Spatial Polygon Triangulation SHI Yong-feng, ZHANG Yu-hao, CHENG Ting, XU Bao-wen, LIN Gang-shan 2019, 40(3): 447-451.  DOI: 10.11996/JG.j.2095-302X.2019030447 Abstract ( 73 )   PDF (888KB) ( 112 )   The problems with the traditional surface segmentation intersection lie in the choice of plane, loss of intersection line and discontinuous intersection line. In view of these problems, this paper proposes a surface segmentation intersection algorithm based on spatial polygon triangulation. The algorithm based on equal depth segmentation avoids the problem of discontinuous intersection line. When the segmentation reaches a certain level, the spatial polygon is used to approximate the surface patch, and the spatial polygon is triangulated. The intersection of the triangular pair is similar to the intersection between the spatial polygons, then the intersection of the spatial polygons is approximated to the intersection of the patches. Finally we get intersection lines between the intersection surfaces. The spatial polygons constructed by the contours of the surface patch are closer to the true shape of the patch, and the approximation accuracy is improved. The triangulation of the spatial polygons improves the accuracy of the intersection, thus reducing the possibility of losing the intersection line. In theory, this algorithm is more accurate than the traditional segmentation method, and the experiment also verifies this conclusion. Related Articles | Metrics

Select

Research on the Interaction Efficiency and Vertigo Sensation of Head Rotation MA Ming-ming1,2, YE Quan2,3, HU Yang2,4, BO Zheng1,2, SUN Dong-biao2,5, WANG Dang-xiao6 2019, 40(3): 452-459.  DOI: 10.11996/JG.j.2095-302X.2019030452 文 Abstract ( 153 )   PDF (897KB) ( 139 )   How to achieve 360-degree observation through the redirection method, and at the same time ensure the user’s higher comfort level and lower sense of vertigo, is an important topic in VR research. In this paper, the interaction efficiency and vertigo perception of the amplifying and guiding head rotation algorithm are studied quantitatively: the users sitting in a couch without movement and rotating to a comfortable angle when wearing a head-mounted display (Pico neo Basic Edition) to view a 360-degree virtual range. Two amplified head rotation algorithms are designed and compared with two existing methods. At the same time, an automatic detection algorithm for guiding the area of interest in head rotation is designed, and the algorithm can dynamically calculate the angle threshold  of the virtual object and select the area of interest in real-time. Based on this, a sine-based guided rotation algorithm is designed.  Through a great deal of experiments, much important information about the experimental comparison results of each algorithm is obtained. Finally, the problem of how to redirect the head rotation in a relatively complicated virtual environment is solved. Related Articles | Metrics

Select

Plant Depth Maps Recovery Based on Target Features CHEN Guo-jun, CHENG Yan, CAO Yue, LI Sheng 2019, 40(3): 460-465.  DOI: 10.11996/JG.j.2095-302X.2019030460 Abstract ( 77 )   PDF (748KB) ( 114 )   Considering the depth images of plants provided by depth camera are incomplete, and common filtering methods cannot inpaint the plant depth images accurately, we propose a plant depth images inpainting method which is based on target features. Firstly, targets of plant color images are segmented by using a color image segmentation algorithm based on color and spatial information, then the outer contour of each target is retrieved, and polygon for each outer contour is fitted. Secondly, the pixels with correct depth value in the depth images are searched to act as sampling points, and meanwhile the leaf maps are normalized. Finally, using spatial fitting method to calculate every target area’s equation to correct the small area’s depth pixels which need to be corrected. In the meantime, support vector machine and spatial transformation are used to get the accurate large area’s depth pixels which need to be corrected. The experiments show that the proposed method achieves better performance for plant depth image inpainting, and protects targets’ edge information. Related Articles | Metrics

Select

Nonintrusive System for Measuring the Amount of Fuel Volume Based on Computer Vision in Aircraft Tank HUANG Wen-feng1, YU Hai-yu2, XUE Jun-xiao2, PAN Jun3, GUO Yi-bo4, XU Ming-liang4 2019, 40(3): 466-472.  DOI: 10.11996/JG.j.2095-302X.2019030466 Abstract ( 81 )   PDF (633KB) ( 117 )   In order to measure the aircraft fuel in any flight attitude, a new aircraft fuel measurement system is proposed. First, the non-contact liquid level sensing technology based on computer vision was used to measure the oil level in the fuel tank. Then we employed the CAD model to model the fuel tank and correct the flight attitude error. Real-time observation and three-dimensional reconstruction of the current liquid level outside the fuel tank are realized, and the fuel in the fuel tank is not required to be energized, thereby avoiding electromagnetic interference and circuit wear problems of the conventional measurement method, and greatly improving the safety and reliability of the fuel system. Accurately measuring the remaining fuel quantity in real time is of great significance to the functions of the aircraft’s fuel supply control, system status monitoring, and fault diagnosis isolation. Compared with the traditional method, the oil quantity measurement method has the advantages of digitization, high intelligence degree, strong anti-interference ability and high reliability. Related Articles | Metrics

Select

Design and Implementation of Interactive Mask Erasing Animation Based on Personalized Customization WANG Hua1, LI Shen-shen1, HE Xiao-yu1, ZHU Fu-bao1, YAO Ni1, XU Ming-liang2 2019, 40(3): 473-480.  DOI: 10.11996/JG.j.2095-302X.2019030473 Abstract ( 61 )   PDF (2446KB) ( 113 )   With the rapid development of virtual reality and mobile internet technologies, many businesses hope to achieve an interactive mask erasing animation through the mobile terminal. However, there are problems with personalized real-time customizations, interactivity and so on. Therefore, we introduce an interactive mask erasing animation method based on personalized customization. Our method includes personalized customization module and interactive mask erasing animation visualization module. The personalized customization module initializes and processes customized data. The interactive mask erasing animation visualization model performs the four tasks of mobile terminal shaking - visual terminal mask erasing - the background data querying and editing. We built a prototype system to test the above method. Simulation results show that it can realize the real-time synchronous erasing dynamic visualization for 2 000 times shaking at mobile terminals. Related Articles | Metrics

Select

An Adaptive Control Method of Running Platform for Personalized Itness YAN Ming1, WU Xin-li1, WANG Rui-rui1, ZHANG Wen1, YANG Wen-zhen1, PAN Zhi-geng2,3 2019, 40(3): 481-488.  DOI: 10.11996/JG.j.2095-302X.2019030481 Abstract ( 74 )   PDF (735KB) ( 119 )   Currently the running platform is difficult to ensure the safety and effectiveness of exercise, and cannot realize the personalized fitness. This paper proposes an adaptive control method of running platform for personalized fitness. To meet the runner's physical health and fitness goals, the appropriate exercise intensity values are calculated to match the runner’s personalized fitness. During the running exercise, the runner's heart rate is measured in real time. The slope and speed of the running platform are automatically adjusted by the practiced heart rate to ensure the effective exercise, which shall be in the scope of his exercise intensity. 40 subjects were tested by this exercise intensity adaptive control method. Their average heart rate ratio in exercise was in the range of (0.678, 0.834), and the standard deviation of heart rate was in the range of (3.156, 6.312). The results show that this control method can stably control the exercise heart rate of the subjects within the optimal fitness heart rate interval, and the heart rate fluctuation is small, which not only ensures the effectiveness of the exercise, but also reduces the possibility of running fitness risk. Thus the effectiveness of this control method for the running platform is verified. Related Articles | Metrics

Select

A Rapid Method of Road Network Modeling for  Intelligent Transportation Systems SHI Min1, SUN Ke1,2, MAO Tian-lu2, ZHENG Ling1 2019, 40(3): 489-496.  DOI: 10.11996/JG.j.2095-302X.2019030489 Abstract ( 97 )   PDF (1325KB) ( 335 )   A simulation road network model for intelligent transportation system has been proposed. Compared with the traditional road network model, this model enhances the geometric expression and topological expression ability. At the same time, it can attach rich traffic attribute data to meet the refined needs of intelligent traffic simulation. On this basis, a road network data migration method is studied, which quickly extracts and constructs a road network for intelligent transportation systems from existing electronic maps. Tests on geometric accuracy, topological completeness and modeling efficiency show that the road network model and its modeling method improve the efficiency of traffic simulation basic road network modeling, reduce the modeling cost, and provide an efficient and reliable road data source for the wide application of intelligent traffic simulation. Related Articles | Metrics

Select

Unified Material Point Method for Realistic Simulation of  Fluid-Solid Interaction SHAO Xu-qiang, ZHENG Kai-xuan, XU Wei-feng, ZHAO Wen-qing 2019, 40(3): 497-502.  DOI: 10.11996/JG.j.2095-302X.2019030497 Abstract ( 100 )   PDF (7828KB) ( 55 )   The simulation of fluid-solid interaction is an important research topic of physically based simulation technology. It is particularly difficult to solve the problems of non-penetration and slip contact at the interface . This paper puts forward a unified material point method (MPM) for realistic simulation of fluid-solid interaction animation. First, we propose an MPM-based method for simulating weakly compressible fluid, and solve the momentum equation of fluid and solid on the unified background grid. Second, we construct multiple background grids in fluid-solid contact regions, and propose a momentum-conserving velocity-position correction scheme to prevent the penetration artifacts and simulate slip contact. The results demonstrate that the proposed method produces stable and realistic fluid-solid coupling animations which can be applied to computer graphics and virtual reality. Related Articles | Metrics

Select

Fine-Grained Image Classification Based on Text and Visual Information YUAN Jian-ping1, CHEN Xiao-long1, CHEN Xian-long1, HE En-jie1, ZHANG Jia-qi2, GAO Yu-dou2 2019, 40(3): 503-512.  DOI: 10.11996/JG.j.2095-302X.2019030503 Abstract ( 109 )   PDF (833KB) ( 180 )   The fine-grained image classification generally only focuses on the partial visual information of image, but in some problems the text information of partial image has a direct relationship with the classification result. By extracting the semantic information of the image text, the image classification effect can be further improved. We comprehensively consider the visual information and local text information of image, and then propose an end-to-end classification model to solve the problem of fine-grained image classification. On the one hand, the deep convolutional neural network is used to obtain the visual features of the image, on the other hand, according to the proposed end-to-end text recognition network, the text information of the image is extracted, and then the visual feature and the text feature are merged by the correlation calculation module and sent to the classification network. Finally, we test the results of our method in the image classification on the public dataset Con-Text, and also verify the end-to-end text recognition network on the SVT dataset, which is better than the previous method. Related Articles | Metrics

Select

A Shadow Ray Traversal Algorithm Based on Multiple-Index Tree LIANG Xiao1, HUANG Yun2 2019, 40(3): 513-518.  DOI: 10.11996/JG.j.2095-302X.2019030513 Abstract ( 92 )   PDF (333KB) ( 121 )   Shadow ray traversal is abig computation bottleneck in ray tracing. However, constructing an efficient tree to cull down redundant intersections is quite difficult. We propose a Multiple-index Tree based on shadow ray traversal algorithm, which adds indexes for nodes to accelerate traversal with acceptable pre-computation. First, since occluded rays try to intersect with primitive, we select primitives with high intersection probability from leaf nodes to store in inner nodes, which aims to stop traversal in upper tree. Second, since un-occluded rays try to find the nearest node, we create adjacency indexes between nodes in bottom tree, and use the indexes to access next node along ray direction directly. During traversal, by exploiting frame coherence, we estimate the occlusion type of rays and use corresponding method to reduce traversal cost. The experimental result suggest that the algorithm can improve traversal performance more than 20% for complex scenes. Even compared with tree reconstruction method, our method outperforms in reducing more intersections and only consumes 21% pre-computation time. Related Articles | Metrics

Select

Image Retrieval Method Based on Perceptual Hash Algorithm  and Bag of Visual Words YANG Wen-juan, WANG Wen-ming, WANG Quan-yu, WANG Jun-jie 2019, 40(3): 519-524.  DOI: 10.11996/JG.j.2095-302X.2019030519 Abstract ( 112 )   PDF (3102KB) ( 129 )    As the existing image retrieval technologies in mobile augmented reality have a low real-time performance caused by long time-consuming, this paper proposes a novel image retrieval method which combines the perceptual hashing and bag of visual word model (BoVW). The method is able to accelerate the search speed with certain accuracy. First, the improved perceptual hashing is used to retrieve a image set in which each image is similar to the current image, which limits the scope of the target. Then a BoVW model is built based on this image set, the BoVW model is used to create a visual vector for each image in the image set and the current image. Finally, hamming distance of the visual vector between the current image and each image in the image set is calculated to finish the image retrieval. The results show that the improvement of our method in accuracy is 3.2% and the retrieval time is reduced by 102.9 ms to the traditional BoVW model algorithm. Our method is able to meet the real-time requirements of image retrieval in mobile augmented reality. Related Articles | Metrics

Select

Facial Expression Simulation Technology for Virtual Avatar YAO Shi-ming1, LI Wei-hao 1, LI Wei-qing 2, SU Zhi-yong1 2019, 40(3): 525-531.  DOI: 10.11996/JG.j.2095-302X.2019030525 Abstract ( 78 )   PDF (6805KB) ( 128 )   In order to realize the offsite visualization interaction function in the electronic sand table environment based on Augmented Reality, a 3D expression simulation technology for virtual avatar is proposed. First, the expressions of remote worker is tracked using a RGB camera, the face feature point data is extracted based on constrained local models (CLM) and the data is sent to local. The interpolation algorithm based on Radial Basis Function is used to calculate the coordinates of vertices of the virtual avatar face mesh, the model is driven to simulate the same facial expression with the other offsite worker; Finally, in order to improve the accuracy and efficiency of the deformation algorithm, a selection method of interpolation control points and a sub-region interpolation method based on the distribution of the human muscle group and the greedy algorithm are proposed, The experimental results show that the algorithm can meet the real-time and realistic requirements of applications. Related Articles | Metrics

Select

Human Posture Recognition Method Based on Few Key Frames Sequence CAI Xing-quan, TU Yu-xin, YU Yu-jie, GAO Yu-feng 2019, 40(3): 532-538.  DOI: 10.11996/JG.j.2095-302X.2019030532 Abstract ( 80 )   PDF (555KB) ( 114 )    This study focuses on the problems that the traditional human posture recognition data acquisition is easily disturbed by environment, and it’s difficult to solve the similarity of human motion postures and the characteristics difference of the human motion executor. This paper proposes a human posture recognition method based on few key frames sequence. Firstly, the original motion sequence is pre-selected. The initial key frame sequence is constructed by taking the extremum of the motion trajectories, and the final key frames sequence is obtained by using frame subtraction algorithm. Then, we built the hidden Markov model for different human postures and trained the model. The Baum-Welch algorithm is used to calculate the initial probability matrix, the confusion matrix and the state transition matrix, and the post-training model is obtained. Finally, the probabilities for each model are achieved by inputting the measured data and applying forward algorithm, and the gestures corresponding to the maximum probability are compared and selected as what is identified. Experiment results show that our method can efficiently select key frames of the original motion sequence, and effectively improve the accuracy of human body gesture recognition. Related Articles | Metrics

Select

A RANSAC-Based Cylindrical Axis Feature Representation for Point Clouds GAO Chun-yan, SHEN Zi-ming, ZHANG Ming-lu, TIAN Ying 2019, 40(3): 539-544.  DOI: 10.11996/JG.j.2095-302X.2019030539 Abstract ( 92 )   PDF (1852KB) ( 131 )   Existing three-dimensional map building algorithms emphasize the accuracy of map building, resulting in low efficiency and high cost of mapping. In order to improve the efficiency of building a map, an improved algorithm is proposed to recognize and extract the cylinder of a landmark object and to construct a simplified map with its axis features as landmarks. Based on random sampling consistency (RANSAC) algorithm, this algorithm generates and matches the cylinder model to be estimated in the point cloud model. By calculating the homography matrix and its error function, the optimum threshold in the iteration process is obtained to get the optimum matching cylinder model and improve the extraction efficiency. Then, the spatial position of the landmark is described by the extracted cylinder axis. Cylindrical radius describes the spatial geometric information of landmarks. Compared with the traditional RANSAC method, the simulation results show that this method can effectively simplify the map and lay the foundation for the subsequent path planning of landmark recognition. Related Articles | Metrics

Select

A Method for Constructing Linear Birational Mappings in Quadrilaterals WU Meng, CHEN Chong, WANG Xu-hui, QIAN Yi-jia 2019, 40(3): 545-548.  DOI: 10.11996/JG.j.2095-302X.2019030545 Abstract ( 77 )   PDF (346KB) ( 120 )    As a special kind of rational mapping, birational mapping can be applied to image warping and domain parametrization of Isogeometric Analysis. By geometric methods, reference [1] constructed a birational mapping on a quadrilateral, but this method needs to compute the weights first and then construct a birational mapping. By the technique of moving lines, a method for the construction of birational map on quadrilateral is proposed. Hence, the computation of weights is not needed here, our method can generate birational map directly. In addition, by choosing different parameter, different birational mappings are calculated. It is easy to verify that the weights of the mapping generated by our method satisfy the weight proportional relation proposed by reference [1]. An example is also provided to illustrate our method. Related Articles | Metrics

Select

Shape Features of a Kind of Trigonometric Bézier Curve WU Rong-jun 2019, 40(3): 549-555.  DOI: 10.11996/JG.j.2095-302X.2019030549 Abstract ( 60 )   PDF (2787KB) ( 115 )    In many applications of geometric modeling, curves of desirable shape should eliminate the unnecessary singularities and inflection points. Therefore, to avoid potential risk in shape design, it is essential to predict and analyze the shape features of parametric curves. In order to quickly determine the shape features of parametric curves, the shape conditions of the parametric curve are simplified due to the homogeneous property of cones, and the cusp conditional cone and two boundary loop conditional cones are obtained for a quadratic trigonometric polynomial Bézier curve characterized with two shape parameters. These three characteristic cones and their tangent planes divide the characteristic space into different characteristic regions. The curve's shape features are completely determined by the distribution region which the characteristic point locates in the characteristic space. It is shown that the shape diagrams obtained by the method based on the theory of envelopes and topological mappings can be derived from characteristic space by virtue of planar slices, which are vertical to one of the axes. Furthermore, the influences of shape parameters on the associated characteristic regions are also discussed. The obtained results enable the user to place control points or choose shape parameters so that the resulting curve is globally or locally convex, possessing wanted singularities or inflection points, or enjoying the desired shape features. Related Articles | Metrics

Select

Research on 3D Gesture Interaction System Based on Leap Motion LIU Yu-xing, WANG Shu-xia, XU Guang-yao, LAN Wang-gui, HE Wei-ping 2019, 40(3): 556-564.  DOI: 10.11996/JG.j.2095-302X.2019030556 Abstract ( 151 )   PDF (1324KB) ( 193 )   The rapid development of gesture recognition and the continuous updating of somatosensory devices provide inspiration for 3D gesture interaction; based on Leap Motion gesture recognition and nearest neighbor algorithms, a 3D gesture interaction system is established. Firstly, the theory of gesture design and the theory of interactive gesture design are studied. Based on this, the gesture function and gesture library are designed, and the gesture library is divided into eight gestures. Secondly, the gesture feature extraction is performed to establish the finger key point model and obtain the angle of the gesture feature; then calculate the gesture recognition efficiency of KNN algorithm and SVM algorithm, and KNN improved algorithm achieves better recognition efficiency. Finally, a 3D interactive system is designed, and the gesture is classified into 4 modules, each module having two gesture tasks; 20 people are gathered to extract 1 600 sets of gesture data, and a data analysis of the total sample joint point mean value is conducted. Design a 3D interactive system module, and import 1 600 sets of gesture data in the 3D interactive system created in Unity3D; the interactive design process, user experience analysis and gesture recognition efficiency statistics are completed according to the customized eight gesture driven virtual hands. Through research, it is found that Leap Motion gesture recognition has high recognition efficiency, and the three-dimensional gesture interaction system is fairly innovative. Related Articles | Metrics

Select

Towards Human-Computer Interface Design for Virtual Micro Cataract Surgical System DAI Sha1,2, SI Wei-xin2,3, QIAN Yin-ling2, ZHENG Rui2, WANG Qiong2, XU Dong-liang1, PENGYan-jun4, HENG Pheng-Ann3,2 2019, 40(3): 565-573.  DOI: 10.11996/JG.j.2095-302X.2019030565 Abstract ( 124 )   PDF (22009KB) ( 82 )   Virtual reality based micro cataract surgical simulator is of great significance for the training of ophthalmic surgery in the microscope vision environment. We have presented a hardware/software environment for a highly immersive virtual micro surgical training system. The hardware part of the system is based on HTC Vive and Geomagic Touch X, and the surgical instrument conversion interfaces are used to build a high-fidelity surgical operating environment. In addition, position-based dynamics in the software part is used to simulate the deformation, puncturing and suturing of corneal soft tissue during surgery, while GPU parallel computing is employed to accelerate the computing for interactive environment. Finally, tests show that this system achieves a more natural human-computer interface for micro cataract surgery, which enables young physicians to obtain a highly immersive real surgical training experience to achieve an effective training effect. Related Articles | Metrics

Select

Knuckle Image Offset Measure Feature Learning Based on Laplace Approximation Gaussian Processes YANG Shi-qiang, QIAO Dan, GONG Lu-qi, LI Xiao-li, LI De-xin 2019, 40(3): 574-582.  DOI: 10.11996/JG.j.2095-302X.2019030574 Abstract ( 64 )   PDF (1248KB) ( 113 )   In man-machine coordinated assembly, the precise finger image feature should be extracted and recognized to accurately present the hand posture. In order to enrich the hand information of co-operator, a multi-classification algorithm based on Laplace approximation Gaussian process is put forward to achieve the knuckle recognition based on the hand image. According to the assumption that the category information is unrelated to each other, the learning of knuckle image mid-level offset measure features is transformed into the learning of random quantities. A multi-class Gaussian process learning algorithm based on Laplace approximation Gaussian process is presented by analyzing the posterior computation in binary multi-class Gaussian field and constructing the positive definite kernel function of knuckle image middle-level random information. The knuckle image training and learning and the image recognition of fixed threshold value are conducted using the mid-level distribution learning and predicting algorithm. The recognition results show that this method is feasible. Related Articles | Metrics

Select

Extraction and Reuse of Pattern Configuration for  Handicrafts Personalized Customization WANG Wei-yi1, LV Jian1, PAN Wei-jie1,ZHAO Hui-liang2, TIAN Qiao-ping1 2019, 40(3): 583-590.  DOI: 10.11996/JG.j.2095-302X.2019030583 Abstract ( 91 )   PDF (9704KB) ( 83 )   In order to solve the problem of low reusability of knowledge during handicrafts personalized customization process, a method for the extraction and reuse of pattern configuration for handicrafts personalized customization is proposed, and here the innovative design of batik pattern is taken as an example. Firstly, the batik knowledge was sorted out to the need of handicrafts personalized customization application, the batik knowledge base and case base were built, and a semantic relationship between the two bases was constructed. Through the retrieval of ontology semantics, a batik case that met the user's needs was recommended. Then, shape context was used to extract the patterns of batik paintings, and topological methods was combined to establish the configuration rules of patterns. Shape grammar based on character encoding was improved in the process of configuration reuse, used to describe the transformation of shapes during pattern filling; thereby the batik pattern reuse for personalized customization was completed. This study develops a batik handicrafts personalized customization system, and tests the method feasibility through an application example. Related Articles | Metrics

Select

Research on the Extracting Algorithm of Dongba Hieroglyphic Feature Curves YANG Yu-ting1, KANG Hou-liang2 2019, 40(3): 591-599.  DOI: 10.11996/JG.j.2095-302X.2019030591 Abstract ( 85 )   PDF (1352KB) ( 239 )   Dongba hieroglyphic is a kind of very primitive picture hieroglyphs. In order to increase the recognition rate of Dongba words, extracting the feature curves of glyphs is the key. At present, the retrieval and recognition of Dongba hieroglyphs still use the existing, universal and mature technology, and there is less analysis and discussion on the Dongba hieroglyphs themselves. Therefore, we analyze the shape and structure of the Dongba hieroglyphs in depth, and give a connected-domain priority marking algorithm based on chain codes. It extends the characteristic of Freeman chain code in edge detection and local segmentation of curves, and satisfies the requirements of contour and structure type of feature curve extraction. We use a large number of experiments and comparison with other classical edge detection algorithms to show that the CDPM algorithm has good versatility, scalability and robustness, and the accuracy reaches 98.2%. It lays a foundation for the retrieval and identification of Dongba hieroglyphs. Related Articles | Metrics

Select

Asphalt Pavement Lane Line Removal Method Based on  Mask R-CNN and Improved Criminisi LAN Zhang-li1, HUANG Tao1, WANG Qing-zhen2, TAN Li-yun2 2019, 40(3): 600-607.  DOI: 10.11996/JG.j.2095-302X.2019030600 Abstract ( 101 )   PDF (15302KB) ( 76 )   In the automatic classification of the disease images of asphalt pavement, there are a great number of images with lane line, which is subject to interference. A method of lane line removal was proposed to reduce its impact on classification. Firstly, the detection model of the lane line region under complex background was trained based on the Mask R-CNN network, and the mask of the lane line region was automatically obtained through the model. Then the mask was used to completely remove all the lane line areas to get the damaged image. Finally, a modified Criminisi image inpainting method was used to fill the damaged image samples. Experiments show that the missed detection rate and the false detection rate are 0.50% and 7.87% respectively with the application of the Mask R-CNN method to detect the road image in 400 different environments. The improved Criminisi method enhances the repair speed by about 4 to 5 times than before under the premise of ensuring the quality of image restoration. Using VGG classification model for comparison verification, the new data set obtained after removing the lane line by the algorithm performs better under the same conditions. Related Articles | Metrics

Select

Research and Application of Faster-RCNN Based M. Tuberculosis Detection Method BIAN Jing-shuai1, LU Jia-pin1, LUO Yue-tong1, ZHANG Min2 2019, 40(3): 608-615.  DOI: 10.11996/JG.j.2095-302X.2019030608 Abstract ( 214 )   PDF (5150KB) ( 179 )   Through sputum-smear staining, mycobacterium tuberculosis can be shown on microscope image, which makes it possible to detect M. tuberculosis on the image for facilitating tuberculosis diagnosis. On the microscope image, M. tuberculosis is characterized with diverse color saturation, various shape, and undistinguishable appearance confused with background, which make it a great challenge for traditional object detection methods. As convolutional neural networks (CNN) has achieved great success in object detection recently, we study CNN-based method, for instance, Faster-RCNN for M. tuberculosis detection. Nevertheless, there are still some problems with CNN-based tuberculosis detection:  a) Size of M. tuberculosis on image is too small, b) Constructing enough accurate labeled data is difficult, and c) Transfer learning does not work for tuberculosis detection. All of those make it hard to apply CNN-based method to M. tuberculosis detection directly. To overcome these problems, we adopt two strategies. We present overlapping sub-image partition strategy for the small-size problem caused by anchor structure which is component of prevalent CNN-based object-detection method. The partition strategy overlappingly partition raw image into sub-images as per a formula presented by us. After partitioning, the proportion of M. tuberculosis on input image of model have been increased, that improving detecting accurate but reducing detecting speed. According to practice, we deem it acceptable. By cooperating with the co-author, 13 261 labeled data of M. tuberculosis have been constructed. Through a series of experiments, it has proved that our method is effective not only in improving detecting accurate and generalization of the model, but also in reducing necessary labeled data. The methods have been integrated into medical inspection products and confirmed to satisfy practical application requirements. Related Articles | Metrics

Select

Investigation on PPT Generation from Mobile Images Taken on Conference, Seminar and Presentation Scenario DU Zhen-long, ZHOU Fei, SHEN Hai-yang, LI Xiao-li 2019, 40(3): 616-622.  DOI: 10.11996/JG.j.2095-302X.2019030616 Abstract ( 94 )   PDF (8756KB) ( 57 )   Smartphones have become a powerful assistant to modern people in terms of taking pictures on the scenarios of conference, seminar and representation, but obtaining information based on these images is discrete, discontinuous, and easily interrupted. Therefore, a method of generating PPTs from mobile pictures is in urgent need, via which we can effectively browse the core content. The paper presents a PPT generation method from the mobile images taken on the scenario of meeting, seminar and representation, including effective region extraction, content area reprojection and content area completion. The effective region is extracted by using the morphological operation with large kernel, highlighting the difference between the content area and the background area, convenient for extracting the effective content area. Based on the corner constraint, the content area is virtually reprojected for generating the regular rectangular area. The content area completion is to remove the obstacles within the content area. In experiments of a variety of scenarios, the proposed method could effectively generate PPT from the images of conference, seminar and presentation. Related Articles | Metrics

Select

Failure Management and Support System of Mechanical Parts Based on Failure Knowledge-Base WU Kang-qing1,2, HUANG Li-ping1,2, LI Bo-shu3, SHI Jin-da4 2019, 40(3): 623-630.  DOI: 10.11996/JG.j.2095-302X.2019030623 Abstract ( 99 )   PDF (1231KB) ( 123 )    In order to solve the difficulty in describing failure knowledge of mechanical parts and the low efficiency of sharing and reuse of failure knowledge among different designers, a failure knowledge base is constructed based on ontology modeling technology. A mechanical parts failure knowledge management & support system (FKMS) is designed and developed using Django-Web module. First, by analyzing knowledge requirement, the function structure of the system is established and the content of failure knowledge is determined. Next a failure knowledge ontology model and its entity relation network are constructed by ontology modeling technology, which ensures the integrity of the knowledge base structure. Then the mapping of knowledge model to database is established based on vertical model, enabling efficient storage and extensible structure. In addition, through failure reason networks based on the ontology mode, a fault reason analysis algorithm with belief propagation is put forward. Finally, the coating spallation of helicopter free-wheel sprag is used as a case in the process of knowledge management and failure analysis, which demonstrates the dynamic management support of the system. Related Articles | Metrics