The hierarchical block boundary element method (HBBEM), which can extract the whole interconnect capacitance matrix with one computation, is of very high efficiency. In analog integrated circuit layout, the feature size varies largely in different layers. According to this, we present an improved HBBEM in this paper, including a new hierarchical partition method of 3D blocks, the nonuniform partition of boundary elements and improved algorithm organization. Numerical results show that the new algorithm is several times faster than the original HBBEM and suitable for the capacitance extraction of real analog integrated circuits, with high accuracy as well.
As one of the important artistic styles of portrait, sketch portrait has wide applications for both digital entertainment and law enforcement. In this paper, an automatic face sketch generation approach is presented by learning from photo-sketch pair examples. Specifically, the relationship between a face photo and its corresponding face sketch is learned on image patch level. By applying this relationship to the input face photo patch, we can infer the output face sketch patch by exploiting some regression techniques such as kNN, the Lasso and so on. Via our local regression model, we can synthesize an appealing sketch portrait from a given face photo in a few minutes. Experiments conducted on CUHK database have shown that our results are more compelling than previous methods especially in two respects: (1) our synthesized sketches preserve more identity information of the original face photo, (2) our synthesized sketches presents more pencil sketch texture.
This paper presents a RSMT algorithm based on gravitation direction, namely G-Tree. It uses the gravitation a point receives from other points to judge its moving direction. Some weights are added to the calculation to reduce the number of bends, which is helpful for via reduction and reliability increment in the routing phase. After the construction, some refinements are taken to reduce bends further more. G-Tree has been implemented and tested. The experimental results show that G-Tree can reduce numbers of bends.
Information security is one of the most important challenges in computer communication. This paper proposed a novel robust watermarking algorithm with BCH error correction coding for the transmission robustness of encrypted images. Besides applying BCH Error correction coding to produce robust marks for the transmission, the scheme utilizing Ant Colony Algorithm (ACA) to pick the image contour as the embedding position for better performance against channel noises. Experiment results illustrate that the proposed scheme can improve the robustness and invisibility of watermarking against attacks, such as JPEG compression, channel Gauss noises interference and cutting operation.
The magnitudes of the endpoint tangent vectors are optimized in the process of Hermite interpolation so that the curvature variation of the optimized geometric Hermite curve is a minimum. The tangent angle constraints guaranteeing an optimized geometric Hermite curve geometrically smooth is got and proved. For the cases in which the given tangent vectors do not satisfy the constraints, new methods for constructing 2-segment and 3-segment composite optimized geometric Hermite curves are presented. Examples have been presented to shown that combination of these new methods and those based on strain energy minimization can get satisfying results in all the tangent angle regions.
The objects with mesostructure surfaces have difficulty in geometric modeling and have complex model structures, which result in low efficiency of shadow rendering algorithms. This paper presents a real-time rendering method for soft shadow based on height gradient map. To realize the real-time rendering, we propose a mesostructure description model, which has the mesostructure height field tied to the surfaces of low-resolution models by using texture coordinates to exhibit complex mesostructure models. We also introduce mesostructure height gradient map (MHGM) to create local apex sets of visible points. In the shadow rendering stage, the detailed shadow is generated by computing the occlusion of gradient-space local apex sets in real time. Our algorithm runs entirely in image space, and uses deferred shading and the graphics pipeline to show the self-shadowing effect of deformable objects in dynamic scenes.
In this paper, we present a high-quality texture-based approach for the visualization of vector fields on surface. It adopts a scheme of streamline enhancement by applying a ID high-pass filter to flow textures along the orthogonal flow direction to enhance the intensity contrast among streamlines and improve the image quality of flow textures. The scheme of edge detection and supplement is used in our approach to modulate the characteristic of particles flow. Our algorithm is image space based and can achieve a real-time rendering performance on a personal computer by using parallel processing ability of current graphics hardware.
We design a pertinence graphical model, combined with domain-specific heuristics among the components of human hand, to describe the hand's 3D structure, kinematics, dynamics and self-occlusion. The modular structure facilitates tracking each hand component (sixteen variables of six degrees of freedom) separately instead of tracking hand configuration of 27 degrees of freedom as a while to reduce the computational complexity. Then, a more efficient belief propagation method embedding continuously adaptive mean shift (CAMSHIFT) algorithm to obtain configuration space (C-space) is proposed. Belief propagation is processed in the feasible C-space to increase the tracking efficiency. The experimental results show that our proposed method can track articulated hand robustly and efficiently under self-occlusion.
Characteristics of tongue pose the most important information for tongue diagnosis. So far, extensive studies have been made on extracting tongue surface features. Meanwhile, the sublingual vein diagnosis, one important part of tongue diagnosis, has received increasing attention. In this paper, a novel image acquisition device specially designed for capturing sublingual vein images is introduced. Different from existing tongue image acquisition devices, monochrome industrial CCD with enhanced near infrared sensitivity is used under near infrared light source. Corresponding segmentation method of sublingual veins for the captured near infrared sublingual images is proposed subsequently. Experimental results reveal that the proposed method did indeed segment the sublingual veins from the near infrared sublingual images with an acceptable degree of accuracy.
Through geomet ric observation , it is found that the nearest point on a curve is a tangent point between the curve and an off set curve of the other curve. Based on this observation , an off set method is presented for computing the minimum distance between two planar algebraic curves. The new method is geomet rically inst ructive , and can be used for computing the minimum distance between an algebraic curve and a parametric one on the same plane. For planar quadratic curves , the degree of the resulting univariate polynomial equation by our method is much lower than that of the equations in previous comparable methods , which may lead to lower computation complexity or higher robustness of the solutions.
An algorithm to construct biarcs that not only match the interpolation requirement of the point positions and tangent vectors but also lie on one side of the constraint line is proposed. For C-type biarc interpolation with the line constraint , firstly twelve inequalities in total are provided according to different position relationships between the constraint line and the interpolation requirement. These inequalities are used to check if biarcs lie on one side of the constraint line. If there exist many biarcs satisfying both the interpolation requirement and the line constraint , the optimal biarc can be obtained by solving a minimum problem. Otherwise , one extra point is added to make it possible to obtain biarcs. For S-type biarc interpolation with the line constraint , some results are also listed.
This paper proposes an optimally weighted linear algorithm to further improve the accuracy of linear algorithm for camera calibration with 1D objects. A new estimator for the relative depths are proposed by optimally weighting their constraint equations only once, and it is shown that non-iterative optimal weighting can ensure similar accuracy to iterative weighting as well as reduced computational cost; The constraint equations on the image of the absolute conic are also optimally weighted only once, where the optimal weights are appropriately simplified to further reduce the computational cost. The experimental results demonstrate that the calibration accuracy of the proposed algorithm is higher than that of the existing weighted linear algorithm, and is comparable to the accuracy of the bundle adjustment algorithm.
A new 2.5D line drawings cartoon model is presented in the paper. Using the model, given two orthographic views of a line drawing cartoon picture, we can generate a set of new views of the cartoon object in these two pictures and simulate the rotation from one view to another by an specific interpolating techniques proposed in the paper. First, all strokes in the two given pictures are extracted. Then, the correspondences of strokes in these two pictures are computed by shape similarities and paired. Third, from two paired strokes, a depth of a plane that the two strokes lie on is acquired. Fourth, by interpolating points of two corresponding strokes and rotating the resultant stroke in 3D space, the stroke viewed from an arbitrary direction between the two given directions was generated. The visibility of interpolated strokes was determined by the painter's algorithm. Finally, a plausible rendering of the cartoon picture from any views between the two given directions could be generated. Comparing to previous methods, the experimental results show that the model is independent of particular line drawing systems, and the rendering effect is much improved.
3D films have become more and more popular due to their exceptional visual quality enhancement in comparison with the traditional 2D ones. However, due to the high cost in 3D film production which severely hampers its fast development, the conversion technology from 2D to 3D becomes a promising way to alleviate the problem. In this paper, the various 2D to 3D film conversion technologies are summarized. More specifically, the classical conversion approaches are first reviewed according to the four kinds of relative motion between camera and scene, then several more advanced ones are discussed, finally some promising future directions are listed.
In view of the obvious shortage of commonly used regional division of gray level-average gray level two-dimensional histogram, an improved two-dimensional histogram based on gray level (average gray level)-gradient and its corresponding regional division method are proposed. The formulas for threshold selection of maximum entropy based on the improved two-dimensional histogram regional division are derived. The particle swarm algorithm is used to search the best threshold. A recursion method is used in iteration to greatly reduce the repeat computations of fitness function. Experimental results show that the proposed algorithm not only achieves a good segmentation quality of uniform regions, accurate borders and robust noise resistances, but also the computation efficiency is promoted twofold compared with the particle swarm algorithm.
The research of depth estimation from a single monocular image is promoted by available massive video data. Under the assumption that photometrically similar images likely have similar depth fields, in this paper we propose a novel 2D-to-3D method based on semantic segmentation and depth transfer to estimate depth information from a single input image. Firstly, semantic segmentation of the scene is performed and the semantic labels are used to guide the depth transfer. Secondly, pixel-to-pixel correspondences between the input image and all candidates are estimated through SIFT flow. Then, each candidate depth is warped by SIFT flow to be a rough approximation of the input's depth map. Finally, depth is assigned to different objects based on semantic labels guided depth fusion. The experimental results on Make3D datasets demonstrate that our algorithm outperforms the existing depth transfer methods where the average log error and relative error were reduced by 0.03 and 0.02 respectively.
In order to strengthen transfer function (TF)'s ability in identifying meaningful features and simplify interactive mode, we present a TF design method based on 2D histogram image segmentation with the guidance of feature difference evaluation. We compute different features to design 2D TFs and use these TFs to separate interesting structures in volume progressively. We also present a feature difference evaluation method to assist choosing features intelligently when designing 2D TFs. By applying these TFs in series to separate interesting structures in volume, we can get a fine separation result which will be highlighted in visualization to perfect rendering result. Besides, we learn from previous work and use normalized cut method to segment 2D histogram image in order to explore volumes hierarchically, which makes our interaction with transfer function more convenient. Some experimental results demonstrate the effectiveness of our method.
In order to explore the efficiency of finite circle method (FCM) and to reduce the circle number as much as possible, three new methods, i.e. the bisection algorithm, the three-step algorithm and the improved three-step algorithm with gap are proposed for the automatic generation of circles in this paper. The first one is to generate the circle for each borderline segment of the component by considering the segment length and the prescribed tolerance value. If this is not the case, the borderline segment will be equally divided into halves until the obtained circle satisfies the tolerance value over the whole borderlines of the component. The basic idea of the last two algorithms aims at covering the domain occupied by the component as much as possible. The circle generation is carried out sequentially from salient angle to the sector formed by the salient angle till the uncovered borderline segments of the component. Particularly, the three-step algorithm with gap allows that the generated circles are distributed with gap along the component contour. Lots of numerical examples are finally tested. It is shown that the three-step algorithm and the improved three-step algorithm with gap can approximate the component with fewer circles. Thus, they can improve considerably the efficiency of multi-components packing and layout optimization.
Critical points such as saddle point are one set of the most important features in 2D vector field visualization methods. A novel clustering based approach is proposed in this paper. First, a set of streamlines are generated, and are clustered into different groups. Conjugated normal vector field and dense matrices are introduced to accelerate the sorting of the streamlines in each group. Thereafter, different post-processes can be performed according to various visualization applications: a streamline simplification that works for each group; a multi-resolution evenly-spaced streamline placement based on streamline proximity; an enhanced texture based visualization based on the weighted matrix. Experiments show that our results are satisfactory and robust.
We present a method to construct a group of isometric and scaling invariant shape descriptors for shape analysis. Firstly, we construct a kernel function based on Biharmonic distance. Then Lebesgue integral is employed to produce a group of isometric shape descriptors. In the following step, Lebesgue measure of the surface is applied to smooth away their scaling factors. Thus we get one group of isometric and scaling invariant shape descriptors on 2D manifold, which reflect some intrinsic properties of shapes. Several experiments are done on the human hand models. The results show that objects can be recognized and distinguished using these shape descriptors with high efficiency.
To alleviate the shortage of 3D media sources, a 2D-3D video conversion method is proposed in this work which is based on layered optimization and integration on the preliminary results from structure from motion (SFM). The main steps include: Firstly, depth cue of scene sparse structures is obtained by a segment-wise SFM approach. Secondly, an extended interactive segmentation algorithm based on GrabCut is used to extract different object layers, and then dense depth generation and optimization are carried out in each layer independently to enhance local visual stereoscopic quality. Thirdly, depth of different layers are integrated considering the visual saliency features to complete 2D-3D video conversion. Experiments show that the proposed method can yield more visually satisfactory results.
The conventional 2D equiscale matrix transform is unable to directly scramble rectangle images. For a rectangle image, it needs to expand it into a square image or divide it into several square images, which brings extra computational cost. To address this problem, this study proposes a 2D bi-scale rectangular mapping for rectangle image scrambling, and it is proved that the proposed 2D bi-scale rectangular mapping is the necessary and sufficient condition for any 2D dual-module mapping to be a one-one mapping. The inverse mapping of the 2D bi-scale rectangular mapping is also proposed by combining two kinds of special inverse mappings of 2D equiscale matrix transform and 2D bi-scale triangular mapping. The experimental results show that the proposed mappings are of validity in scrambling any rectangle images, of low cost in scrambling and recovering rectangle image, and of robustness to erasing, cropping and compressing attacks. In contrast to the conventional scrambling methods, the proposed mappings no longer need to compute the minimum recovering period, and have the least iteration cost and the largest transform matrix generation space.
To improve the performance of routing algorithms of network-on-chip, leveraging more global information of network status is natural and necessary. However, from which region to obtain global information and how nodes in the region weight are not fully considered in current adaptive routing algorithms. This paper firstly analyzes the properties of 2D mesh topology and then proposes a novel routing algorithm leveraging global information. This algorithm calculates the weight of each node in critical area and stores in the central control module, then figures out the routing table and updates the routing table of each router in regular time. The experimental results of SPLASH-2 traces show there are 32% and 18% reduction of average packet latency in comparison with the traditional dimension order routing, and local greedy routing algorithm respectively.
According to the psychology analysis of human visual perceiving, people tend to pay more attention to the foreground than the background in watching videos. In this paper, a novel 2D to 3D conversion method based on segmenting the foreground from the background is proposed. Firstly, the moving objects are detected by background model and refined by attention model so as to extract the foreground objects in interests. Secondly, every object in the background is found by machine learning method, and the depth of the background object is estimated based on the pin-hole camera model. Thirdly, the depth of the foreground is estimated by its location and the frame coherence. Finally, the depth of foreground and background are integrated. The experiments showed that our algorithm can be easily applied to many scenes and the result of depth estimation reflects the location of each object in the real world exactly.
The existing methods have difficulties in fast generation of solid texture similar to a given sample. Considering the typical marble and wood textures, this paper presents an efficient technique for fast generation of solid texture from 2D sample. According to the spatial distribution of the texture, the movement path for 2D sample is designed in 3D space. All the pixels of the sample are then used to colorize the 3D space along their trajectory so that the solid texture is obtained. The quintic Catmull-Rom spline is designed for noise interpolation to construct turbulence, which is adopted to disturb pixel trajectory to simulate the irregularities of the natural texture. By placing triangles vertical to their longest side and assembling triangles in descending order according to the heights, a novel texture atlas generation algorithm is presented to store mesh texture obtained from the solid texture space. Experimental results show that the algorithm only takes a few seconds to generate highly realistic solid texture on mesh surfaces, which is very similar to the sample texture.
2D shape blending has been widely used in 2-D character animation, pattern matching, and geometric modeling. Previous algorithms mostly tend to use local geometric elements such as edge lengths, angles and areas, to associate the regions on the two shapes which look alike, but oblivious to the intrinsic feature structures of the shapes. This paper presents a new method that blends planar shapes in terms of their intrinsic feature structures. Firstly, the source and target shapes are decomposed into several pairs of features, respectively, which are further associated via a semi-automatic approach that is based on user's heuristic. Secondly, a multi-level feature structure is defined, representing the global positions and orientations, local orientations, and local details of the shape features. During the shape blending, the feature structure is interpolated from its source to target configurations at each level. Lastly, intermediate shapes are derived from the interpolated results of all levels of the feature structure. Experimental results show that the method can effectively avoid shape distortion and preserve local shape features, as well as generate smooth, natural and visually pleasing effects.
A comprehensive tooth shape modeling scheme is presented to reconstruct the missing part of each tooth due to teeth adhesion on the 3D dental model. The adhesion region between two adjacent teeth on the 3D dental model is first detected by exploiting the minimum principal curvature information. Then, the adhesion region is cut off and the tooth profile opening caused by the removal of the adhesion region is filled with triangular patches adopting a local optimization scheme, it is then refined to form an intermediate restoration surface patch. The final restoration result is obtained by minimizing the surface energy to reflect the bio-characteristics of the single tooth. Experimental results demonstrate that the missing profile shape of a single tooth on the 3D dental model can be restored faithfully, thus meeting the requirements of dentistry, oral surgery & medicine.
It is often a non-intuitive and tedious task to develop mixed reality (MR) applications due to the difficulty of accurate 3D registration and the inconvenience of virtual 3D model creation. In this paper, we propose a novel integrated mixed-reality 3D authoring tool. In order to achieve accurate registration with high geometric consistency, both natural features and artificial markers are employed and combined for robust structure-from-motion computation. In order to efficiently construct virtual objects with accurate registration, the proposed authoring tool is integrated into a commercial package of 3D digital content creation. The experimental results demonstrate that the proposed authoring tool can achieve the objective of efficient virtual model construction and accurate registration.
3D facial animation is a highly important but difficult task in computer graphics area. An effective approach on editing and synthesis of 3D facial animation is proposed. By the approach, users are enabled to assign constraints of facial movement through simple interaction in 2D screen space. A prior probabilistic model is trained by our system using facial animation data. With such a prior model, we propagate the user constraints to the whole 3D facial mesh to generate complete and lifelike facial expressions. By using Isomap learning algorithm, we model the knowledge of 3D facial animation. Incorporating the user specified key frames as end points, a smooth geodesic curve on the high dimensional space of facial animation data is approximated to produce novel animation sequence. Experimental results show that by the proposed method we can intuitively control the synthesis of facial animation and produce vivid expressional animation.
The key point in a MPEG-4 based facial animation system is the construction of the FaceDefTables, We proposed an automatic method to locate feature points on 3D facial models and then construct FaceDefTables for arbitrary facial models based on the FaceDefTables for the standard facial model. At last, we implemented the facial animation system that can animate arbitrary facial model based on the FAPs realistically.
To realize the fast collision detection between objects, a sphere approximation representation for 3D models is proposed in this paper. It firstly presents a novel 3D model approximation method with inner spheres based on three-dimensional model voxelization. It also reduces the number of inner spheres and enhances the sphere connectivity. Then this approach improves the inner sphere clustering based on the geodesic distance. It obtains meaningful clustering results which can be applied in many applications such as the structure construction of hierarchical inner sphere-trees and the semantic segmentation of 3D models. Experimental results demonstrate the feasibility and effectiveness of this method.
3D facial expression synthesis has been an important and challenging task in the field of computer animation. Inspired by the fact that facial expressions distribute on a nonlinear manifold, we propose an approach for 3D facial expression synthesis based on nonlinear co-learning. Firstly, 3D facial expressions with the same attribute are projected onto an identical low dimensional representation according to the theory of nonlinear co-learning, by means of unsupervised regression. Secondly, based on the low dimensional representations of 3D faces, reconstruction operations are needed to synthesize expressions for given 3D faces, and to retarget expressions based on given expression samples. The proposed approach is able to handle noisy/incomplete input faces, and generate intact expressional faces. Experimental results show that the proposed 3D facial expression synthesis outperforms the existing methods both in quality and in efficiency.
To reuse CAD models more efficiently, a new method for seeking the common reusable partial structures from a large amount of CAD models is proposed. Firstly, CAD models are represented by the attributed feature adjacent graph (AFAG). Then, the frequent sub-graph mining algorithm is employed to detect the common reusable partial structures. The partial structures are obtained through candidate generation, candidate pruning, frequency counting and post-processing. Experimental results show the effectiveness of this approach for the extraction of the reusable partial structures in the CAD models with different shapes. Besides, its efficiency meets the requirement of engineering application by applying multiple optimization methods in extraction process.
In recent years, computer vision and machine learning have been heavily applied in the research field of 3D human animation. Video/image based motion capture, digital character/scene modeling, interactive character controlling and motion synthesis techniques are all based on the computer vision theory. And machine learning has played an important role in the research of 3D motion data reuse and intelligent character animation and achieved some wonderful results. A survey on the computer vision and machine learning techniques used in 3D human animation is given in this paper, also with some discussion to the future work provided.