To appear in the ACM SIGGRAPH conference proceedings High Dynamic Range Display Systems

合集下载

金属薄片的自适应变形

金属薄片的自适应变形孙晓鹏;洪靖惠【摘要】针对金属薄片通过弯折、冲压成型等变形过程,提出关于金属薄片变形的快速、有效的自适应仿真框架.首先基于金属薄片材料的几何及物理特性,修改自适应重新网格化的加劲肋约束;并对网格的最佳尺寸约束进行改进,以模拟金属的高延展性;最后给出简单高效的金属薄片自适应变形仿真的系统框架.实验结果表明,该框架可适用于任何金属薄壳.%Aiming at the deformation process of sheet metal by bending and stamping, this paper proposes a fast and effective adaptive simulation framework for sheet metal deformation.Firstly, based on the geometrical and physical properties of the metal foil material, the adaptive re-meshed the stiffening rib restraint is modified.Then, the optimal size constraint of the mesh is modified to simulate the high ductility of the metal.Finally, a simple and efficient system framework for adaptive deformation simulation of sheet metal is presented.The experimental results show that the framework can be applied to any metal shell.【期刊名称】《计算机应用与软件》【年(卷),期】2017(034)005【总页数】6页(P68-72,85)【关键词】金属薄片;塑性变形;拉伸弯曲;自适应网格;仿真【作者】孙晓鹏;洪靖惠【作者单位】辽宁师范大学计算机与信息技术学院辽宁大连 116000;辽宁师范大学计算机与信息技术学院辽宁大连 116000【正文语种】中文【中图分类】TP39对金属薄片施加外力进行弯折，然后解除，金属薄片恢复形变的程度非常小。

Tree Vertex Splitting Problem-

CS 668 - Spring ‘06Student Research Presentations(RB 122)Date/Time Presenter TitleMonday, April 10 4:00 pm Fabien Poulard Utility of random generated graphs tooptimize peer-to-peer networksMonday, April 10 4:30 pm Todd Chaffins Performance Comparison of VertexColoring Algorithms on CPUs and GPUsWednesday, April 12 4:00 pm James Haberly Graph-theoretic image processingtechniquesWednesday, April 124:30 pmMahbub Majumder Graceful labeling of treesMonday, April 17 4:00 pm Hsiaoying Su Path-finding - An application of graphtheory in computer gamesMonday, April 17 4:30 pm Mandeep SinghAtwalTree Vertex Splitting Problem -Applications to Distribution NetworksWednesday, April 19 4:00 pm Bryan Ritz Analysis of Techniques Used in DrawingGraphs with Subgraphs (with Case Study) See the following pages for abstracts.Monday, April 104:00 pmUtility of random generated graphs to optimize peer-to-peer networksFabien PoulardThe first peer-to-peer networks emerged with the Gnutella protocol in the early 2000, thanks to a team of developers at Nullsoft. When tested in a lab, the protocol was able to manage some hundreds nodes; but when released, it was quickly used by thousands of users, creating as much nodes and revealing the weaknesses of its conception. Since then the peer-to-peer has evolved and some others protocols have appeared, fixing partially the weaknesses of Gnutella. However, those protocols do not really implement a real peer-to-peer network, but a kind of topologically hybrid network sharing characteristics of the decentralized topology and some characteristics from other topologies (most likely hierarchy and centralized).As at the origins the peer-to-peer was developed to share music files (Nullsoft develops a widely used mp3 player product) “illegally”, there was no more research to extend the capabilities and exploit the peer-to-peer networks. However, nowadays, peer-to-peer is considered as a possible topology for a lot of networks, and especially for the mobile device applications. So, improving the algorithms at the origin of Gnutella becomes industrially interesting.Much progress has been achieved in the last few years, especially due to graph theory. The main issues to resolve are about preserving the connectivity of the graph when adding or removing a peer and in making the graph evolve towards an expander graph to optimize the exchanges between clients. However, the huge constraint is to make that happen with a minimum number of transactions considering that each manipulation of the network implies latencies and loss of packets that will have to be resent. Another issue is the fault tolerance, as each node can fail due to the extreme heterogeneity of the network, it must always be possible to find another path from one node to another.The different techniques I will try to cover during the presentation are based on the random transformation of regular graphs and the de Bruijn graphs. We will see how they can be applied to the original Gnutella to improve its efficiency without losing its decentralized characteristics.References• Peer-to-peer Networks based on Random Transformations of Connected Regular Undirected Graphs, P. Mahlmann, C. Schindelhauer, ACM July 2005• Graph-Theoretic Analysis of Structured Peer-to-Peer Systems : Routing Distances and Fault Resilience, D. Loguinov, J. Casas, X. Wang, IEEE/ACM Transactions on Networking • Distributed Construction of Random Expander Networks, C. Law, K. Siu, IEEE INFOCOM• The diameter of random massive graphs, L. Lu, Proceedings of the twelfth annual ACMSIAM symposium on Discrete algorithms• Generating Random Regular Graphs, J.H. Kim, V.H. Vu, Proceedings of the thirty-fifth annual ACM symposium on Theory of computing• On the Fundamental Tradeoffs Between Routing Table Size and Network Diameter in Peer-to-peer Networks, J. Xu, IEEE INFOCOMMonday, April 104:30 pmPerformance Comparison of Vertex Coloring Algorithms on CPUs and GPUsTodd ChaffinsCurrent high-end graphics processing units (GPUs) have evolved from dump co-processors with fixed functionality into highly capable stream processors. [Purcell, et al. 2002] In the ever-growing games industry there is a demand to be able to create more realistic graphics and effects. The current trend is to create these realistic graphics and effects through the use of graphics code known as shaders [Harris, et al. 2003]. Shaders are small pieces of code which determine how graphics are rendered on the screen. These shaders and their adoption in games require the graphics card manufacturers to create faster and more programmable GPUs. This pressure will continue and as such the speed and programmability of GPUs are set to continue to increase as time goes on. [Christen 2005]This increase in processing power and the parallel nature of GPUs has led to the adoption of the GPU for general purpose computations outside of the realm of graphics. While the GPU has made vaster architectural changes over recent years CPUs have also made advances. Aside from the common increases associated with CPUs (clock speed and cache), CPUs have moved to be more parallel with a dual-core architecture. With parallel vertex-coloring algorithms [Kale, et al. 1995] the question is raised as to which approach will yield the fastest results when performing vertex-coloring: CPU based, GPU based, or a hybrid CPU/GPU approach. This research seeks to implement, instrument, and measure the performance of these approaches in a quantitative manner and evaluate the economy of these approaches.References:[Purcell, et al. 2002] Purcell, T.J., Buck, I., Mark, W.R. and Hanrahan, P. Ray Tracing on Programmable Graphics Hardware. In Proceedings of SIGGRAPH 2002, ACM / ACM Press. 2002.[Christen 2005] Christen M.: Ray Tracing on GPU. Diploma thesis, University of Applied Sciences Basel, Switzerland, 2005.[Kale, et al. 1995] L. V. Kale, B. H. Richards, and T. D. Allen. Efficient parallel graph coloring with prioritization. In Lecture Notes in Computer Science, volume 1068, pages 190{208. Springer-Verlag, August 1995.[Harris, et al. 2003] Mark Harris, Greg James, Physically-Based Simulation on Graphics Hardware, GameDevelopers Conference, 2003.Wednesday, April 124:00 pmGraph-theoretic image processing techniquesJames HaberlyThe aim of my proposed topic would be to present an exposition on Graph Theoretic approaches and algorithms as they’re being researched for use in image processing and machine vision. Some example problems in the area of image processing and machine vision are; computational complexity, object recognition, object measurement, image segmentation, edge detection, noise detection and filtering, line, arc and other feature detection, image coding and compression.The project will be focused on presenting how graph theoretic algorithms with examples such as the Prim and Kruskal algorithms for minimum spanning trees, Dijkstra’s and Dial's shortest path and graph theoretic Euclidean distance mapping techniques are being examined for problem solving in the image processing and machine vision fields.I do not want to yet limit the scope of this proposal by proposing an exposition on any one paper or problem since I’m just beginning the research phase of the project. A couple specific areas that may become the main focus of the project are:1. Improvements in computational speed using graph-theoretic image processing techniques [1].2. Object recognition using a graph theoretical approach [2].The image processing and machine vision industry is a fast growing and exciting field. I’m looking forward to researching the topic further.References:[1] “Faster Graph-Theoretic Image Processing via Small-World and Quadtree Topologies” Leo Grady and Eric L. Schwartz Dept. of Imaging & Visualization, Siemens Corp. Res. Inc., Princeton, NJ, USA; This paper appears in: Computer Vision and Pattern Recognition, 2004. CVPR 2004. Proceedings of the 2004 IEEE Computer Society Conference on Publication Date: 27 June-2 July 2004Volume: 2, On page(s): II-360- II-365 Vol.2[2] "Color Invariant Object Recognition using Entropic Graphs" Jan C. van Gemert, Gertjan J. Burghouts, Frank J. Seinstra, Jan-Mark Geusebroek Intelligent Systems Lab Amsterdam, Informatics Institute, University of Amsterdam, Kruislaan 403, 1098 SJ Amsterdam, The Netherlands.[3] “Graph-Theoretical Methods in Computer Vision” Ali Shokoufandeh1 and Sven Dickinson2 G.B. Khosrovshahi et al. (Eds.): Theoretical Aspects of Computer Science, LNCS 2292, pp. 148–174, 2002.Wednesday, April 124:30 pmGraceful labeling of TreesMahbub MajumderA tree T with n vertices is said to be gracefully labeled if its vertices are labeled with the integers [1..n] such that the edges, when labeled with the difference between their endpoint vertex labels, are uniquely labeled with the integers [1..n-1]. If T can be gracefully labeled, it is called a “graceful tree”.The concept of graceful labeling of trees and graphs was introduced by Rosa (1967). The term “graceful labeling” was invented by Golomb (Golomb 1972). The Graceful Tree Conjecture states that all trees are graceful. There have been over 670 papers to date on various graph labeling methods and issues (Gallian 2005). So far, no proof of the truth or falsity of the conjecture has been found. Even though the conjecture is open, some partial results have been proved (Gallian 2005).My motivation in pursuing this project came from its nature and the study many people have put into it. My aim with this project work is to1.Find out current works and results.2.Study and understand the condition of gracefulness3.If possible, add some ideas in graceful LabelingReferences:Gallian J. A., A Dynamic Survey of Graph Labeling (2005), Electronic Journal of Combinatorics.Golomb S.W. How to number a graph. In R.C. Read, editor, Graph Theory and Computing, pages 23-37. Academic Press, 1972.Rosa A., On certain valuations of the vertices of a graph, Theory of Graphs (Internat. Symposium, Rome, July 1966), Gordon and Breach, N.Y. and Dunod Paris (1967) 349-355.Monday, April 174:00 pmPath-finding - An application of graph theory in computer gamesHsiaoying SuGraph theory is widely used in solving and presenting computer games. One of the most common applications is path-finding. Path-finding algorithms grant agents in the virtual world the ability to consciously find their own way around the land. They can also be used in real life to find driving directions, such as the service offered by many popular web sites. The project is going to focus on the game implementation.Path-finding algorithms are usually, but not only, used in computer games catalogued as role playing games (RPGs). Programmers build a virtual world for the games. The characters, or called autonomous agents, have certain level of artificial intelligence. The simplest way to present their intelligence is to find their own paths moving around the world without hitting a tree or going through a wall. The successful implementation of path-finding is important to the artificial intelligence performance of a game.According to the numbers of sources and destinations, path-finding algorithms can be roughly divided into three categories: single source, single pair, and all pairs. In single source algorithms, the path from one node to all the others is required. However, single pair algorithms take a specific source and destination. Only one path is required in response. The all pairs algorithm returns the shortest paths from every node to all other nodes.This project plans to explore different path-finding algorithms and their complexity. Furthermore, coding these algorithms in JEdit and practically experiment their efficiency. In single source algorithms, Dijkstra’s Algorithm and Bellman-Ford-Moore, which deals with negative arc-lengths graph, would be discussed. Then, the most popular algorithm A* (A star) would be presented to implement single-pair shortest path finding. To find out all-pairs shortest paths, the algorithms described above could be used in a naïve but inefficient way. The project then would seek out whether there exist more efficient algorithms to solve the problem.Path-finding related topics have been discussed on more application than research. However, there are still some interesting studies going on. One of a recent research is about solving incoherent behavior of multiple units in a cluttered environment. Another one is to discuss an open problem of emulating the rich complexity of real pedestrians in urban environment. The two papers are listed below as references.References•Kamphuis A., Overmars M. H.: Motion planning: Finding Paths for Coherent Groups using Clearance. In Eurographics/ACM SIGGRAPH Symposium on Computer Animation (2004).•Shao W., Terzopoulos D.: Artificial intelligence for animation: Autonomous pedestrians. In Eurographics/ACM SIGGRAPH Symposium on Computer Animation (2005).Monday, April 174:30 pmTree Vertex Splitting Problem - Applications to Distribution NetworksMandeep Singh AtwalIn an Ethernet network, the number of connections (taps) and their intervening distances are limiting factors [BN90]. Repeaters are used to regenerate the signal every 500 meters or so [BN90]. If these repeaters were not used, “standing waves” (additive reflections) would distort the signal and cause errors [BN90]. Because collision detection depends partly on timing, only five 500-meter segments and four repeaters can be placed in series before the propagation delay becomes longer than the maximum allowed time period for detecting a collision [BN90].Directed acyclic graphs (dags) or directed trees can be used to model such interconnection networks. Each edge of such a tree is labeled with a real number called its weight. Trees with edge weights are called weighted trees. Nodes or vertices in the tree correspond to receiving stations and edges correspond to transmission lines [HSR98]. Each edge weight corresponds to the delay in traversing that edge [HSR98]. However, as stated above, the network may not be able to tolerate losses in signal strength beyond a certain level.In places where the loss exceeds the tolerance level, repeaters have to be placed. Given a network and a loss tolerance level Tree Vertex Splitting Problem is to determine an optimal placement of repeaters.RESEARCH OBJECTIVESThe proposed research aims at the study of Tree Vertex Splitting Problem (TVSP). Designing the algorithms for TVSP and analyzing in terms of the computing time and space requirements. The most efficient algorithm can then possibly be implemented inC++/Java.However, it is not an objective to implement the algorithm for the proposed study. REFERENCES[BN90] Barry Nance, Network Programming in C, QUE Corporation 1990,ISBN: 0-88022-569-6, page # 23.[HSR98] Ellis Horowitz, Sartaj Sahni, Sanguthevar Rajasekaran, Fundamentals of Computer Algorithms, Galgotia Publications’ Pvt. Ltd. 1998,ISBN: 81-7515-257-5, page # 203.[PRS98] Doowon Paik, Sudhakar Reddy, Sartaj Sahni, Vertex Splitting In Dags And Application To Partial Scan Designs And Lossy Circuits, International Journal of Foundations of Computer Science, 1998.[ME93] Matthias Mayer, Fikret Ercal, Genetic Algorithms for Vertex Splitting in DAGs, Proceedings of the 5th International Conference on Genetic Algorithms, 1993, ISBN: 1-55860-299-2[SR96] Stephanie Forrest, Genetic Algorithms, ACM Computing Surveys, Vol. 28, No. 1, March 1996.Wednesday, April 194:00 pmAnalysis of Techniques Used in Drawing Graphs with Subgraphs(with Case Study)Bryan RitzIn the paper “Drawing Graphs Within Graphs” [5] by Paul Holleis, Thomas Zimmermann, and Daniel Gmach, the authors present methods for helping to reduce complexity of large and complicated graphs and subgraphs. The methods of finding an optimal layout of subgraphs and a summary graph, of the use of connection sets, and of the use of motifs are all combined into an approach for emphasizing subgraphs within graphs. This paper will attempt to evaluate the worthiness of these methods through examination of sources used in the paper and by applying the methods to a case study in the form of a complex graph (displayed using JEdit).References:1. F. J. Brandenburg. Graph clustering 1: Cycles of cliques. In Proceedings of the Graph Drawing 1997, volume 1353 of Lecture Notes in Computer Science, Berlin, Germany, 1997. Springer.2. G. Di Battista, P. Eades, R. Tamassia, and I. G. Tollis. Graph Drawing: Algorithms for the Visualization of Graphs. Prentice-Hall, Englewood Cliffs, N.J., 1999.3. T. M. J. Fruchterman and E. M. Reingold. Graph drawing by force-directed placement. Software-Practice and Experience, 21(11):1129-1164, 1991.4. T. Kamada and S. Kawai. An algorithm for drawing general undirected graphs. Information Processing Letters, 31(1):7-15, 1989.5. P. Holleis, T. Zimmermann, D. Gmach. Drawing Graphs Within Graphs. Journal of Graph Algorithms and Applications, 9(1):7-18, 2005.Case Study will use TouchGraph as applied to .1) Go to /TGGoogleBrowser.html2) Enter “” without the quotes in the text field and hit enter.。

基于学习的超分辨率技术

—193—基于学习的超分辨率技术郑丽贤，何小海，吴炜，杨晓敏，陈默（四川大学电子信息学院图像信息研究所，成都 610064）摘要：基于学习的超分辨率算法使用一个图像训练集来产生一个学习模型，运用该模型为输入的低分辨率图像创建更多的高频信息，获得比基于重建算法更好的结果。

该文介绍了基于学习的超分辨率技术的相关工作、理论基础和主要算法，提出基于学习的超分辨率算法中仍需解决的关键问题，展望其在未来的研究发展方向。

关键词：超分辨率；马尔可夫随机场；图像金字塔Learning-based Super-resolution TechniqueZHENG Li-xian, HE Xiao-hai, WU Wei, YANG Xiao-min, CHEN Mo(Image Information Institute, College of Electronics and Information Engineering, Sichuan University, Chengdu 610064)【Abstract 】Learning-based super-resolution technique predicts the high-resolution images from the input low-resolution ones, through learning from a training set which consists of a large number of other high-resolution images. And the results are better than the reconstruction based super-resolution algorithms. The related work, theory and algorithms of learning-based super-resolution are illustrated. The crucial problems which need to be resolved in further work are proposed. Directions of future research are pointed. 【Key words 】super-resolution; Markov random field; image pyramid计算机工程Computer Engineering 第34卷第5期Vol.34 No.5 2008年3月March 2008·人工智能及识别技术·文章编号：1000—3428(2008)05—0193—03文献标识码：A中图分类号：TP391.41目前的图像超分辨率技术主要分为两类：基于重建的超分辨率技术和基于学习的超分辨率技术。

科技节学术科技作品-1118

昭通市第一中学明达分校教学学术论文作品名称：探索高中物理实验的重要性作者：张涛职务：中学物理教师探索高中物理实验的重要性张涛昭通市第一中学明达分校摘要：基于色彩传递的灰度图像的彩色化是将一幅彩色图像的颜色特征传递给另一幅灰度图像,使灰度目标图像具有与源彩色图像相似的颜色。

本文采用Welsh的将彩色参考图像上的颜色转移到被着色的灰度图像上的思想，在Lαβ颜色空间和YCbCr色彩空间，分别对七副不同的灰度图像进行了彩色化处理，并讨论了不同颜色空间对不同颜色比重的图像着色效果的影响。

关键词：物理实验作用重要性比重1 引言物理学是一门基础自然科学，它所研究的是物质的基本结构、最普遍的相互作用、最一般的运动规律以及所使用的实验手段和思维方法。

随着人类对物质世界认识的深入，物理学一方面带动了科学和技术的发展；另一方面推动了文化、经济和社会的发展。

经典物理学奠定了两次工业革命的基础；近代物理学推动了信息技术、新材料技术、新能源技术、航空航天技术、生物技术等的迅速发展，继而推动了人类社会的变化。

物理学又是现代医学和药学的基础学科之一，它的理论和实验方法被广泛地应用于现代医学和药学中，并且正在积极地推动着现代医学和药学的发展。

然而实验又是物理学这门自然科学的基础，回顾物理学产生和发展的过程，可以看出物理实验自始至终都占有极其重要的地位。

无论是经典物理学的建立和发展，还是现代物理学的进展都与物理实验息息相关。

可以预言，未来物理学的发展一定离不开物理实验。

可见物理学从根本上说是一门实验科学。

物理学中的每一个概念的建立，每一个规律的发现都有着坚实的实验基础。

任何物理规律的发现和物理理论的建立都必须以实验为基础，并经受实验的严格检验。

物理学就是在理论与实验相互推动下不断向前发展的。

因此，物理教学也必须遵循物理学的规律，强调理论与实验相结合，把实验课和理论课放在同等重要的地位。

对学生而言，在物理实验中，“观察现象、进行演示和学生实验，能够使学生对物理事实获得具体的、明确的认识，这是理解概念和规律的必要的基础。

基于交错反激的光伏控制器及其控制策略的研究

0 引言微型逆变器是指一种独立光伏组件并网发电系统的功率变换单元，通常的功率等级为100-300W。

反激式变换器常因其结构简单、自带高频隔离、并且开关管驱动容易，因此常应用于微型逆变器。

其中在较低功率的两级式微型逆变器中常使用传统的反激式变换器，但是由于其变压器铁芯磁状态在最大直流成分下，需要铁心开较大的气隙，因此其又有难以达到较大功率的缺点，通常应用在不超过100瓦的场合，并且由于其输出电流断续使输出纹波较大，会对后级逆变带来较大的谐波[1]。

因此，本文提出将交错反激式变换器应用在两级式微型逆变器中前级的光伏控制器中，使其具有输出功率较传统反激式变换器的微型逆变器相比翻倍、输出电流脉动小、容易做MPPT控制等优点[2]。

同时使用不对称模糊控制进行最大功率跟踪，可以快速、稳定地实现最大功率点跟踪。

本文将详细分析该变换器使用在两级式微型逆变器中的工作状态，讨论其具体是如何增大输出功率、减小电流输出纹波、减小直流母线输出电压纹波以及进行最大功率跟踪，最后通过实验加以验证。

1 原理与设计1.1 电路设计及原理分析1.1.1 主电路设计图1为两级式微型逆变器的交错反激变换器电路原理图，该电路初级由两个开关管1S、2S一个输入去耦电容in C 和两个反激变压器1T，2T组成。

它的次级由两个整流二极管1D，2D和一个滤波及功率解耦电容busC，这里的滤波电容同时又作为两级式微型逆变器的直流母线功率去耦电容，以保证直流母线上电压的稳定，其中1S和2S交错导通，构成交错并联反激电路，将太阳能电池板输出的20V-45V的电压变换为380V-420V，提供给后级逆变器产生220V的并网电压。

circuit structure is simple.However,it is difficult to apply in place of above 100 watts because of the power is limited by the transformer magnetic state.Therefore,I put forward using the Interleaved-Flyback-Converter in the two stages MI.It has some advantages compared with Single-Flyback-Converter such as double output power,smaller output current ripple and smaller dc bus voltage ripple.what＇s more,it has the same advantages as Single-Flyback-Converter such as simple circuit structure and easy to track maximum power point.Finally, I use matlab/simulink simulation,using asymmetric fuzzy control for maximum power point tracking（MPPT）.In conclusion，This circuit can be used in the MI of 200w, and can track the maximum power point within 0.0005s under the condition of irradiance changes.Keywords: Interleaved Flyback; Two stage micro inverter; Maximum power tracking; Asymmetric fuzzy control图2 占空比为25%和75%时交错并联反激电路电流连续模式的工作波形因此，本文针对开关管控制信号占空比小于50%和大于50%工作波形分别以25%和75%为例，结合图1进行工作状态的详细分析。

喜羊羊

NPR Today•“Stylized Rendering Techniques for Scalable Real-Time 3D Animation”, Adam Lake, Carl Marshall, Mark Harris and Marc Blackstein, NPAR 2001•"Computer-Generated Pen-and-Ink Illustration of Trees", Oliver Deussen and Thomas Strothotte, SIGGRAPH 2000•"Sketchy Drawings", Marc Nienhaus and Jurgen Dollner, AfriGraph 2004 led路灯、无极灯、led景观灯 Stylized RenderingTechniques for Scalable Real-Time 3D AnimationPaper OverviewLeo Chao•Published–NPAR 2001•(Non-Photorealistic Animation and Rendering)–Authors•Adam Lake•Carl Marshall•Mark Harris•Marc Blackstein•Graphics Algorithms and 3D Technologies Group –Intel Architecture Labs•UNC –Chapel Hill•Prior Work–No novel techniques, adapts from prior works to hardware•Motion Lines–Maic Masuch. “Speedlines: Depicting Motion in Motionless Pictures.”ACM SIGGRAPH 99 Technical sketch.–Siu Chi Hsu and Irene H. H. Lee. "Drawing and Animation using Skeletal Strokes". In Proceedings of ACM SIGGRAPH 94, pages 109-118. 1994.•Silhouette Edges–Lee Markosian, Michael Kowalski, Samuel Trychi, Lubomir Bourdev, Daniel Goldstein, and John Hughes. Real-Time Nonphotorealistic Rendering. InProceedings of ACM SIGGRAPH 97, pages 113-122. 1997.•Cartoon ShadingContributions•Hardware Implementation of NPR Techniques–Stylistic Shading•Cartoon Shading•Pencil Sketch Shading–Stylistic Inking•Silhouette Edge Detection•Stylized Silhouette Rendering–Motion Lines•Fundamentals–Preprocess colour textures–Render as normal–Except:•Modify lighting/shading calculations•May require texture coordinate processing •“Painter”–Modified Shading Step –Cartoon Shading–Pencil Sketch Shading•Cartoon Shading–Reducing detail aids comprehension –No Smooth Shading•Color Reduction•Hard Transitions•Variants SimpleALGORITHM Cartoon ShadePreprocess:1.Calculate the illuminated diffuse color for each material:2. Calculate the shadowed diffuse color:3. For each material, create a texture with two colors-u=1 is coloured C i -u=0 is coloured C s ml m l m g i d d a a a a C ⨯+⨯+⨯=ml m g s a a a a C ⨯+⨯=C i Vertex ColorCs Shadow Diffuse Color a l Light Ambientd l Light Diffuse a m Material Ambientd m Material DiffuseALGORITHM Cartoon ShadeRuntime:1.Calculate the one-dimensional texture coordinate at each vertex2. Render the model using a standard graphics API -Lighting disabled-Enable texture mapping }0,{n L Max L normalized light vector•Pencil Sketch Shading–Extends cartoon shading to 2D textures–Extends cartoon shading to multiple textures –Multiple textures to represent light density–Associate texture coordinates•Project texture from the viewport onto the model•Fundamentals–Identify “important”edges•Silhouette•Border•Crease–Modify drawing of these edges •“Inker”–Modified Line Drawing –Visually important edge detection –Stylized edge drawing•Visually important edge detection –Silhouettes•Front-Face and Back-Face share an edge –Borders•Edge shared by two materials•Unshared polygon edge–Creases•Large dihedral angle between polygonsALGORITHM SEDPreprocess:1.Create edge-list structure with Edge Type flagsa.Create a unique edge list using a hash table.b.Edge Hash Function: sum the two vertexindicesc.Border Flag-Edges with only one neighboring face-Two neighboring faces with differentmaterialsd.Set crease flags-Non-deformable geometry-Dihedral angle of two faces > thresholdALGORITHM SEDRuntime:1.For deformable meshes:a.Calculate face normalsb.Detect crease edges from face normalsc.Set Crease Edge flags2. Detect silhouette edgesa.•Stylized “Important”Lines –Emphasis on these lines–Option 1: Straight Line•Fast/Easy–Option 2: Texture Maps•“Artistic”Style•Curvature driven “strokes”•Involves a graph driven search•This is a HACKMotion Lines•Gives the appearance of motion•For fast moving objects•Simple algorithm•Not much overhead–One circular buffer to maintain position data•Claim: Maintain Real-Time Speeds –30 fps, in 2001•Tests:–Integrated with a level of detail scheme–Altered model complexity to maintain FPSFuture Work•Inker–Better “artistic”lines•Painter–Pencil sketch “doesn’t look right”–Follow surface gradient perhaps •Incorporate other NPR techniques –Those used by artists–Doesn’t say which would amenableQuestions?•ques·tion(kw s ch n)n.–An expression of inquiry that invites or callsfor a reply.•heck·le (h k l)tr.v.–To try to embarrass and annoy (someonespeaking or performing in public) by questions, gibes, or objections“Computer-Generated Pen-and-Ink Illustration of Trees”(SIGGRAPH 2000)Oliver DeussenThomas StrothotteMotivation•The paper presents a method for creating pen-and-ink-like illustrations from detailed geometric models.•Motivated by the need for illustrative representations of plant life in architectural and landscape renderings.•We want to be able to produce images similar to this:Related Work•“Cartoon Tree” –Alvay Ray Smith–Use small disks to represent bunches of leaves •Reeves and Blau–Use structured particle system to represent realistic trees.•Kowalski et al.–Two step process1.Render shaded primitives that approximate foliage2.Overlay graftals on rendered image using Salisbury’s linedrawing methodRelated Work Examples•Kowalski’s Graftal Trees:•Reeves “particle forest”:•Smith’s “cartoon tree”:Traditional Illustration of Trees•The Skeleton is drawn up to the second level of branching.•Foliage is broken up into three parts:1.The directly lit canopy –represented by outlines and scarce innerdetail2.The half shadow –greater detail provides an overall darker greylevel3.The shaded area –achieved by thicker lines or drawing leaves inblack•Artists will general use a combination of these to draw the leaves. Rarely do they use all three.•The tree skeleton is represented by generalizedcylinders.•Analytical Silhouette Algorithms–Markosian et al.–Rakar and Cohen–Depth Differences•The skeleton is shaded with crosshatch marks. Marks could be placed using the algorithm outlined by Salisbury et al.–The authors actually use a variant of Floyd-Steinbergdiffusion that places strokes when the accumulated greyvalue is above a threshold.•The xfrog modeling system can generate models with 200,000 leaves on a tree. We want to pair this down to 10,000 or so.•Leaves are represented abstractly by particles: a location and an orientation vector.•The goal is to illustrate groups of with a set of strokes.•Idea: Render particles as primitives and use depth buffer information to determine which leaves are spatially close and outline them.Depth Differences•Initially, each particle is rendered with a view-facing primitive, such asa disk.•Basic Algorithm:–For each pixel compute the maximum positive difference between it and its neighbors•If the distance is above the given threshold, mark as an edge.Depth Differences Continued…•The threshold and primitive size can be varied to achieve different effects:Primitive size: 0.15Primitive size: 0.70Threshold: 1000Threshold: 2000•The threshold can be varied usingother buffers, achieving greater detailin areas that results in a shadowingeffect:Abstract Drawing Primitives•More realistic leaves can be achieved by using the particles’ orientation vectors to interpolate between a set of polygons that represent leaves from different viewsLevel-of-Abstraction•Two Approaches1.Convert raw z-buffer values from non-linear, projected values to linear valuesusing formula below and vary threshold according to depth:d: depth value between 0 and 1d0: minimal depth value (e.g. 0)d1: maximal depth value (e.g. 65535)z0: depth value of near clipping planez1: depth value of far clipping plane 2.Take advantage of the non-linearity to automatically generate more abstractrepresentations.–Geometry farther away will bepushed into fewer values in the z-buffer.– A tree in the background willrepresented by fewer strokes thanthe same tree in the foreground.Level of Abstraction Continued…•The level-of-abstraction can be furthered by scaling the primitives so that the amount of screen space they occupy is independent of their depth.r: scaling factord: desired screen spaces: actual screen spaceNo Scaling ScaledSoftware Implementation1.Calculate depth differences and mark pixels above threshold.•In the interactive mode this bitmap is directly used and blended with other geometry.2.Generate shadows in software. Large number of isolated surfacesprecludes hardware implementation.3.Vectorize bitmap using least squares fitting or primitive identificationmethod.•Draw lines using spline interpolation•Apply line stylesResultsFuture Work•Apply cross-hatch shading to the leaves•Create a continuous level-of-detail algorithm that stays performance while getting rid of artifacts that come with discrete representations.•Apply new styles and color for cartoons.•Questions?Sketchy DrawingsMarc Niehaus Jürgen Döllner University of Potsdam University of PotsdamSketchy Drawings•Sketchiness shows unfinished state •Useful for architectural and product designPrevious Work•G-Buffers (Saito & Takahashi 1990)•Edge enhancement (Nienhaus and Döllner 2003)•Loose and Sketchy filter (Curtis 1998)•Artistic Strokes (Northrup & Markosian 2000)Algorithm Overview•Obtain edge and shading data •Perturb this data to get sketchy look •Problem: It needs to be done in real timeG-Buffers•Need to access in real time•Depth/Color/Alpha/Stencil Buffer on graphics card •Textures•Current hardware allows multiple textures •Manipulate with fragment and pixel shadersEdge Map•Need to find edges for most NPR algorithms•Store three types of edges–Silhouette–Crease–Border•See Edge Enhancement (Nienhaus & Döllner 2003)for more detailsShade Map•Render unlit 3D geometry to capture flat colors•Store results in screen aligned texture instead of color bufferAdding Uncertainty•Apply uncertainty function to edge and shade maps •Store function in texture•Use Perlin function to get frame to frame consistency •Multiply uncertainty values with 2x2 matrix•Use different matrix for edge and shade maps to get a more sketchy feelEnd Result•Use shaders to combine results into texture •Render texture on screen-aligned quad。

1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。

ToappearintheACMSIGGRAPHconferenceproceedingsHighDynamicRangeDisplaySystemsHelgeSeetzen1,2,WolfgangHeidrich2,WolfgangStuerzlinger3,GregWard1,LorneWhitehead2,MatthewTrentacoste2,AbhijeetGhosh2,AndrejsVorozcovs3

1)SunnybrookTechnologies,2)TheUniversityofBritishColumbia,3)YorkUniversity

Figure1:Fromlefttoright:ourprojector-baseddisplayshowinganHDRimage;ourLED-basedHDRdisplayshowingadiscreteandasmoothintensityramp(thetophalfofthediscreterampandthebottomhalfofthesmoothramphaveeachbeencoveredbya1%transparentﬁltertoillustratehighluminancecontentontheleftsideoftheimage,whichcannotbecapturedbythecamera);acolor-codedoriginalHDRimage;HDRphotographtakenoffthescreenofourprojector-basedsystem;HDRphotographtakenoffaconventionalmonitordisplayingthetone-mappedimage.

AbstractThedynamicrangeofmanyreal-worldenvironmentsexceedsthecapabilitiesofcurrentdisplaytechnologybyseveralordersofmag-nitude.Inthispaperwediscussthedesignoftwodifferentdisplaysystemsthatarecapableofdisplayingimageswithadynamicrangemuchmoresimilartothatencounteredintherealworld.TheﬁrstdisplaysystemisbasedonacombinationofanLCDpanelandaDLPprojector,andcanbebuiltfromoff-the-shelfcomponents.Whilethisdesignisfeasibleinalabsetting,theseconddisplaysystem,whichreliesonacustom-builtLEDpanelinsteadoftheprojector,ismoresuitableforusualofﬁceworkspacesandcommer-cialapplications.Wedescribethedesignofbothsystemsaswellasthesoftwareissuesthatarise.Wealsodiscusstheadvantagesanddisadvantagesofthetwodesignsandpotentialapplicationsforbothsystems.Keywords:Hardware–NovelDisplayTechnologies;Rendering–PerceptuallyBasedRendering;ImageandVideoProcessing–ImageProcessing;MethodsandApplication–SignalProcessing;Hardware–FramebufferAlgorithms.CRCategories:B.4.2[INPUT/OUTPUTANDDATACOMMU-NICATIONS]:Input/OutputDevices—Imagedisplay;I.3.3[COM-PUTERGRAPHICS]:Picture/ImageGeneration—Displayalgo-rithms;I.3.4[COMPUTERGRAPHICS]:GraphicsUtilities—Devicedrivers;I.4.0[IMAGEPROCESSINGANDCOMPUTERVISION]:General—Imagedisplays.1IntroductionInthepastfewyears,thelimiteddynamicrangeofbothimagingdevicesanddisplayshasreceivedalotofattentioninthecomputergraphicscommunity.Algorithmshavebeendevelopedforcaptur-ingbothphotographs[MannandPicard1994;DebevecandMa-lik1997;Robertsonetal.1999;MitsunagaandNayar1999]andvideos[Kangetal.2003]withextendeddynamicrange.Simultaneously,tonemappingoperatorshavealsobeendevel-opedforcompressingthedynamicrangesothattheimagescanbedisplayedonthefamiliar8bit/channeldisplayswithtypicalcon-trastratiosofabout300:1,includinganyconventionalCathodeRayTube(CRT),LiquidCrystal(LCD),andprojector-baseddis-play.Whilethesetonemappingoperators(e.g.[Schlick1994;Lar-sonetal.1997;TumblinandTurk1999;DurandandDorsey2002]amongothers)allowfordisplayinghigh-dynamic-range(HDR)im-agesinarecognizableandevenaestheticallypleasingway,nobodywouldconfuseaphotographrenderedinthisfashionwith,say,watchingthesamescenethroughawindow.Thedynamicrangeofconventionaldisplaysissimplyinsufﬁcienttocreatetheopticalsensationofwatchingarealsunsetordrivingacarintooncomingtrafﬁcatnight.Notethatthisisnotjustanissueoftopintensity:simplyincreasingthebrightnessofaconventionaldisplaywouldwashoutthedarktonesandturnthemintoamediumgray.Whatisneededisasigniﬁcantexpansionofthecontrastordynamicrangeofthedisplay.InthispaperwedescribetwoalternativedesignsforHDRdis-playsystems.Wehavebuiltprototypesofboth,anddiscussboththeopticaldesign,andrelatedsoftwareissuessuchasdisplaycali-brationandtherenderingofHDRimagesonbothdisplays.BothdisplaysystemsarebasedonthefundamentalideaofusinganLCDpanelasanopticalﬁlterofprogrammabletransparencytomodulateahighintensitybutlowresolutionimagefromaseconddisplay.Forexample,assumewehaveanydisplaywithacontrastrangeofc1:1betweenthedarkestandthebrightestintensitypro-

1ToappearintheACMSIGGRAPHconferenceproceedingsduciblebythatdisplay.IfwenowputanLCDpanelwithacontrastratioofc2:1infrontoftheﬁrstone,thenthe(theoretical)contrastofthecombinedsystemis(c1·c2):1.Inpractice,theﬁrstdisplayneedstobeabletoproduceaveryhighintensityimage,becausecolorLCDpanelsonlyhaveatransparencyofabout3-8%,evenwhenswitchedto‘white’,sothatmostenergyisactuallyabsorbed.AnotherreasonforusingadisplaywithaveryhighbaseintensityisthatalotoftheHDRimageswewouldliketoshowhave,infact,verybrightregionsinthem.Basedonthisprinciple,wehavederivedtwoalternativedesignsforHDRdisplays.Intheﬁrstdesign(Section4),avideoprojectorbasedonDigitalLightProjector(DLP)technologyservesasthebasedisplay.Inthisversion,wedirectlyfocustheprojectorontothebackoftheLCDpanel.Sincetheilluminatedareaismuchsmallerthanduringregularuseofaprojector,thelightdensityisdramaticallyimproved,yieldingthehightopintensitiesthatweareaimingfor.Whilethisdesignworkswellinalaboratorysetting,ithasseveraldrawbacksthatrestrictitsuseforawiderclassofapplications.Inparticular,thesearealargeformfactor,signiﬁcantpowerconsumptionandheatdevelopment,aswellascalibrationissues.Toovercometheseissues,wehavedevisedaseconddesign(Sec-tion5),inwhichtheprojectorisreplacedwithalow-resolutionarrayofultra-brightLEDs.TheintensityofeveryLEDcanbepro-grammedindividually,yieldingalowresolutionversionofthede-siredimage.Highfrequencyfeaturesareintroducedbyattachingahigh-resolutionLCDpaneltothefrontofthisLEDarray,andadjustingitstransparencyaccordingly.Thisdesignmakesuseofresultsfrompsychophysics,whichshowthatveryhighcontrast,al-thoughimportantonaglobalscale,cannotbeperceivedbyhumansathighspatialfrequencies(seeSection3).Thetwodisplayswedevelopedhavedynamicrangeswellbe-yond50,000:1,andamaximumintensityof2700cd/m2and8500cd/m2,respectively.Thiscomparestoatypicaldynamicrangeofabout300:1,andamaximumintensityofabout300cd/m2foratypicaldesktopdisplay.Thedesignofbothsystemsandtheirappropriaterenderingalgo-rithms,aswellastheadvantagesanddisadvantagesaredetailedinSections4and5.InSection6wediscusspossibleapplicationsforourdisplaytechnologybeforeweconcludewithsomeremarksandfutureresearchdirectionsinSection7.2RelatedWorkTheclassofimageprocessingtechniquesforcopingwiththedis-crepancybetweenrealworldluminancesandthosethatﬁtwithinthelimitedgamutofaconventionaloutputdeviceiscollectivelycalledtone-mapping.TumblinandRushmeier[1993]introducedthisconcepttocomputergraphics,thoughtheirearlyworkdidnotaddressdynamicrangelimitationsperse.Theﬁrsttone-mappingoperatortotackledynamicrangereductionwasChiuetal.[1993],whousedaspatiallyvaryingexposurerampovertheimage.How-ever,thisapproachledtodisturbing“reversegradients”typicallyseenashalosaroundlightsources.LaterworkbyLarsonetal.[1997]returnedtoaglobaloperatorfordynamicrangereductionbasedonhistogramadjustmenttoavoidtheseartifacts,withlocalvariationstosimulatedisabilityglareduetohighcontrastbound-ariesinascene.Pattanaiketal.[1998]developedwhatsomere-searchersconsidertheultimatestillimageoperatorbasedonthehumanvisualsystem,incorporatingcoloradaptation,localcon-trast,anddynamicrange.However,eventhisoperatorexhibitedsomereverse-gradienteffectsnearhighcontrastboundariesduetoitslocalspatialﬁlters,leadingotherresearcherstotakeadifferentapproach.Thebasicchallengeforaspatiallyvaryingtone-mappingopera-toristhatitneedstoreducetheglobalcontrastofanimagewithoutaffectingthelocalcontrasttowhichthehumanvisualsystem(HVS)issensitive.Toaccomplishthis,anoperatormustsegmenttheHDRimage,eitherexplicitlyorimplicitly,intoregionstheHVSdoesnotcorrelateduringdynamicrangereduction.Theﬁrstresearcherstosuccessfullyaccomplishthisinanautomatictone-mappingwereTumblinandTurk[1999]withtheirLCISoperator.However,LCISsometimesproducesodd-lookingimages,whichbearlittlerela-tionorresemblancetotheoriginalscenebrightnesses.Morere-centoperatorsbyAshikhmin[2002],Fattaletal.[2002],Reinhardetal.[2002],Durand&Dorsey[2002],andChoudhury&Tum-blin[2003],aremuchmoresuccessfulinseparatingcontrastdiffer-encesthatmattertovisionfromthosethatdonot.Regrettably,noneofthesetechniqueshasbeenvalidatedorsup-portedbypsychophysicalresearch,sotheresultingimagesremainastheonlyevidencethatthesemethodshaveanyrealvalidityforre-producingourexperienceofanHDRsceneonalowdynamicrangedevice.Mostofthemethodscontainfreeparametersthatmustbesetbytheuserbasedonpreference.Usually,littlecanbesaidaboutthevisualimpactthattheseparametershaveontheimageintermsofvisibility,contrast,brightness,orhumanvisualresponseingen-eral.HDRdisplaysoffertwobeneﬁtsinthisarea.Theﬁrst,immedi-atebeneﬁtistoresearchers,whountilnowhavehadnomeansforthecontrolleddisplayofdynamicHDRimageryintheirstudies.OurHDRdisplaysarealreadyhelpingresearcherstotestouttheirhypothesesregardingtheeffectsoftone-mappingonHDRscenes.ThelongertermbeneﬁtwillbefeltwhenHDRdisplayspenetratetheprofessionalandeventuallytheconsumermarkets,reducingoreliminatingtheneedfortone-mapping.Althoughweexpecttone-mappingtocontinuetobearequirementformanytypesofout-put,suchashardcopy,theavailabilityofHDRdisplayswilllikelyreducetheneedfordynamicrangecompressionformanycriticalapplicationsintheyearstocome.