A QoS Controller for Adaptive Streaming of 3D Triangular Scenes ZhiQuan Cheng and ShiYao Jin PDL Lab., National University of Defense Technology Changsha City, Hunan Province, P.R. China [email protected], [email protected]

Abstract. With the development of network, it’s a common requirement to perceptually access the 3D scenes at a remote server. This demand has not been filled by software standards and their implementations, leaving a wide potential for technical innovations. In the paper, we present a QoS (Quality of Service) controller for remote rendering of 3D contents, aiming at higher real-time performance just as rendering local 3D scenes. The model of our QoS controller is defined by a high-level heuristic function, which has integrated important perceptual factors. And then, guided by the QoS controller, a hybrid transcoding algorithm is developed, which adaptively partitions each 3D model into different modality: the metadata (its compressed base mesh with texture) and a series of refiner meshes. Consequently, the 3D scenes can be adaptively and progressively transmitted to a requesting client, and be fluently displayed at end user’s window in a constant frame rate. Experimental results show the soundness of the proposed controller and algorithm guided by the 3D controller. Keywords: Remote Rendering, Streaming, 3D QoS, VRML, and Adaptation.

1 Introduction Remote visualization of large, dynamic and interactive 3D scenes has been a very important research field since the advent of digital geometry processing technologies. However， this demand has not been filled by software standards and their implementations, leaving a wide potential for technical innovations. Advances in streaming and signal processing technology are necessary to enrich the perception and performance of 3D objects in distributed applications [1,2], e.g. in collaborative environments or applications for remote rendering digital geometry museum. Especially, one important requirement of these applications is that the interactive frame rate should be high enough and should be constant in order to give the user a smooth navigation in the virtual environment. In our previous work [3], we have proposed the Adaptive Progressive Remote Rendering Architecture (APRR) framework, a service-oriented architecture, for progressive delivery and adaptive rendering of 3D contents. Based on the ideas and lessons learned during the previous development, we find that we should and must successfully take human perception factors into adaptive streaming of 3D models. In this paper, a 3D QoS controller is presented in the client/server framework, and K.-c. Hui et al. (Eds.): Edutainment 2007, LNCS 4469, pp. 572–583, 2007. © Springer-Verlag Berlin Heidelberg 2007

A QoS Controller for Adaptive Streaming of 3D Triangular Scenes

573

integrated in both client and server side. The controller guarantees the server to adaptively transmit the best quality of multi-resolution 3D objects in such a way that a minimal overall quality degradation over the scene is obtained. While at the end-user system, the user terminal is refreshed at a constant specific frame rate, by making speed-accuracy trading-off under resource constraints, based on the QoS model. Besides these, we also propose a hybrid transcoding for adaptive transmission of a triangular mesh, which combine the metadata, its basic mesh with a mapping texture, and a series of optimal refined hierarchical levels of detail. In the following paragraphs, we will articulate the related works in section 2. By examining the graphics pipeline in remote rendering of 3D object, section 3 identifies the location of 3D QoS controller, and then describes the QoS framework including a QoS architecture, 3D QoS model and its heuristic function. Section 4 presents a hybrid coding method for adaptive streaming of 3D scenes, and explains how the QoS controller avails to the execution of the streaming algorithm. Experimental results and discussion are given in section 5. Finally, we make a conclusion in section 6.

2 Related Work Remote rendering of 3D models is a hot and important research area, and lots of researches have tackled the problem. However, implementing an adaptive system for delivering 3D models over networks is not trivial, effective researches should be investigated: New pioneering designs and solutions for coding, compression and streaming/broadcasting technologies for 3D models and multimedia data as well as technologies and applications for distribution frameworks, networking technologies for efficient and robust data communication mechanisms on current and next generation networks need to be developed. On the one hand, Hesina and Schmalstieg [4] advocated the distributed framework of remote rendering for the first time. Afterwards, many infrastructures have been development from the reasonable design and development of a logical structure [3,5,6] to the complex scene organization and management at the server [7,8,9]. Especially, the methods based on VRML/X3D has been thoroughly focused by presenting various coding techniques [10,11,12,13,14,15]. On the other hand, 3D compression [16] has been a perdurable branch of data compression, aiming at the compact representation of 3D models, and a lot of excellent progressive algorithms [17] have been proposed. Existing 3D compression algorithms use both techniques adapted from the 1D and 2D cases (like wavelets, entropy coding, and predictive coding), and completely different approaches that take advantage of the properties of 3D surfaces (like vertex degree [18], geometry image [19], and subdivision surfaces [20]). For a more complete survey of techniques in 3D mesh compression, readers are referred to recent papers [21,22]. In addition to the points already discussed, several aspects outside the graphics domain should be further researched. Firstly, at the networking level, error resilience [23,24,25] is important if lossy transmission protocols like User Datagram Protocol (UDP) are used. Secondly, security [26] is paramount to the integration of the adaptive system into the real world, as it constitutes a major concern for all networked graphics applications. Thirdly, a more elaborate pre-processing [27] scheme including

574

Z. Cheng and S. Jin

a more sophisticated perceptual model that accounts for inter component dependencies and methods for seamless integration of different representations is necessary. Finally, statistical representation [28,29] of the geometry models has also been researched for efficient transmission and rendering. In general， each component is independently considered in the previous, which may lead to combinations of different representations in a single image. However, in the paper, we attempt at integrating the perceptual adaptation and real-time preservation function into the progressive framework by a 3D QoS controller.

3 QoS Framework 3.1 Graphics in a Networked World Figure 1 shows a simple abstraction of the graphics pipeline, which includes the phases in the remote rendering. The rendering pipeline describes the way that geometric data takes from modeling to the final image. In this theoretical conception, the local data, held in the main memory of the local client site, does the actual rendering, while the full geometry database is held at the remote server site.

Fig. 1. The abstract graphics pipeline of remote rendering of 3D models

1) Geometry Modeling. This stage is independent to later actual rendering method, and can be performed off-line. Note that adaptive streaming mechanism must be pre-computed to be able to perform level of detail rendering in later stages. 2) Global Scene Traversal and Simple Chosen. The stage of the pipeline includes two jobs. One is traversal of the display scene or database in immediate or retained mode, the other is trival accept/rejection classification of a 3D model by identifying whether it is wholly inside or outside of the view volume of a client.

A QoS Controller for Adaptive Streaming of 3D Triangular Scenes

575

3) The Net. The networking stage transmits the required data to the user client site, based on the demand request for particular pieces of geometric data. 4) Local Scene. The local geometry cache holds local part/whole copy of geometric data from the global geometry database. The main challenge is to stream its content at all times appropriate to the portion of the virtual environment visible to the observer. All the techniques described in this paper aim at improving performance so that this problem can be greatly alleviated. 5) Local Rendering. The rendering process that maps a model to an image on screen is referred to traditional graphics pipeline [30]: local scene traversal, polygon processing, and pixel processing. Each primitive encountered during traversal is passed through the remainder of the local rendering pipeline. The polygon processing does sequent vertex transformations to each primitive, which include model transformation, lighting calculation, viewing transformation and clipping. And the pixel processing is also called rasterization, which is composed of three subprocesses: visible-surface determination, scan conversion and shading. To optimize the stages of the remote rendering pipeline, some knowledge discussed earlier are exploited in the paper. In the infrastructure layer, the APRR framework is made more efficient and real-time performance by adding perceptual adaptive function based on a 3D QoS controller, discussed later in the section. And in the data layer, a transcoding algorithm is proposed in section 4, to adaptively transmission of the progressive meshes. 3.2 APRR with QoS Controller APRR is a service-oriented architecture for 3D content delivery and client rendering, based on three-tier framework. And its prototype has been designed and implemented for network streaming and remote interaction of complex 3D battlefield environment. To enhance the real-time interactive function of the APRR, we introduce a new 3D Quality of Service (QoS) controller in the APRR (Fig. 2) to make speed-accuracy trading-off under resource constraints. Consequently, the display on the client terminal can be rendered in constant frame rate, guided by QoS function that is a perceptual demanddriven geometry transmission strategy. In order to provide QoS at the end user terminal, the middle service tier should negotiate a QoS contract with the network and with the user client. In the following, we would focus on the QoS controller, since the others have been clearly explained in the original APRR [3]. At the server side, the relevant model becomes an adaptive progressive model, after which passes through adaptation generation and management service modular. Similar to [6], the server QoS controller estimates its performance characteristics that are synthesized by T, Q, and I parameters required by each client, where T is the estimated time that APRR deliver model from the server to the client, Q is computed based on the later QoS model and defines the quality how closely a rendering of this representation approximates the rendering of the full-resolution data for each client, and I is the level of interaction. The adaptive management mechanism, guided by the QoS controller, selects the best continuous level available for each model, which fits into the budget, and progressively delivers them to the client in binary format.

576

Z. Cheng and S. Jin

User Interface Graphics Engine Rendering Visible Geomtry View Frustum & Occlusion Culling QoS Adaptation Controller Re-selection

Global Scene Retieval

QoS Controller

Adaptation Generation

Global 3D Scene DB

Adaptation Management

EIS Tier

Network

Data Sets

User Interface Graphics Engine Rendering Visible Geomtry & View Culling

Scene Traversal

Receiver binary

Sender Receiver

Delivery Mechanism

Middle Service Tier

Local Scene Client

Fig. 2. APRR logical architecture with QoS controller: Each color represents an independent “model-view-controller” design pattern modular. The orange points out view modular, the blue denotes on model modular, and green presents controller modular. And the new appended QoS controller and its control information are saliently highlighted in the red color, although they should be masked by the green.

At the client side, the user interface allows the user to specify a desired interactive frame rate and to interact with the 3D scene. The QoS controller makes use of the resource estimator to estimate the processing time for each object based on high-level QoS parameters of the object and platform specific parameters. And then, The function of the QoS controller is to maintain a constant specified interactive frame rate by executing the control loop of Fig. 3, i.e., selecting for each viewpoint the best quality of visible object portions, taking the constraints on the resources into account. Viewpoint Changes ?

Yes

Visibility culling

Benefit/cost update

Rendering

Quality selection

No

Fig. 3. Constant frame rate control at the client

From the before discussion, it’s obvious that the 3D QoS controllers in the client and server has the same structure and QoS model, except that they have different resource constraints. 3.3 QoS Model In order to estimate accurately the streaming level L of each object O under the given resource constraints C(u) for a client user U, it is important to have an accurate 3D QoS model at the viewpoint V. Therefore, we adapt Ngoc et. al.’s method [5] by defining the QoS model as follows: Q(O, L,V , U ) = β (O) * S (O ) *cos(α ) * PSNR(O, L,V )

A QoS Controller for Adaptive Streaming of 3D Triangular Scenes

577

Where Q(O,L,V,U) is the quality of object O, rendered at streaming level L and viewpoint V, which represents the amount of perception contribution of the object to the overall user’s scene. It depends on the size of the object S(O) (in % screen coverage), the sematic meaning β (O) of the object in the scene, the viewing angle α from the user viewpoint to the object, and the quality metric PSNR(O,L,V) approximately computed as the optimization algorithm in [5]. Note that β (O) is the only viewpoint independent parameter. The other factors have to be recalculated whenever the user view volume changes.

4 Adaptive Streaming of 3D Meshes 4.1 A Hybrid Transcoding Algorithm

Generally, the idea of adaptive progressive graphics is to include such methods into a single system that allows optimal combinations to be selected and applied, depending on the specific requirement of each client user. Once each client’s information has been obtained and quantified, the server can accommodate multiple representations of the same model and select optimal ones with respect to given limitation. And then, the challenge lies in extracting the most meaningful part of the data for the client that cannot afford to receive it all. For a 3D model, we propose a hybrid transcoding algorithm to considerably improve the quality and efficiency of its transmission in distributed graphics environments. Enlighten by the idea of texture mapping progressive meshes [19], the hybrid method partitions the compressed model into parts: a series of refiner meshes, and its metadata that is the compressed basic mesh with texture. The single texture of the metadata can be used to texture all approximation in a progressive mesh sequence. z The connectivity and geometry properties of the model are compressed by our earlier anchor-based valence-driven progressive algorithm [31], which has improved the original progressive degree-driven compression algorithm [18]. To get the metadata, we adapt our modified algorithm [31] in two ways. Firstly, to preserve the conformance with the texture-mapping, a set of anchors, locating nearest to the salient texture-mapping position, are selected from the input mesh surface vertices in a feature-sensitive fashion and preserved in the base mesh. Note that adjacent faces on the surface are generally not adjacent in the texture, which is first observed by Soucy et. al. [33]. Therefore, if we don’t use the anchors to record the main texture mapping positions, we cannot use this texture map anymore. Secondly, a multi-granularity quantization method in geometry encoding is presented in [31], which allows making better of the dynamic range (different number of bits in normal/tangential components) than with a fixed-bit representation. Consequently, we achieve the basic mesh of the model and a continuous set of approximating meshes. z The Texture coordinate mapping and value is encoded and decoded in the way, described as Isenburg and Snoeyink [32].

578

Z. Cheng and S. Jin

Now we can turn on the metadata of the model. At the server, if the Oriented Bounding Box (OBB) of the model is touched by the view volume of a client user, it would be directly accepted and downloaded in a whole. And Figure 4 has shown that it faithfully captures the detail at all levels of the streaming mesh sequence. Markv warship

M1a1 tank

58 triangles

40 triangles

35 triangles

140 triangles

100 triangles

60 triangles

354 triangles

279 triangles

156 triangles

700 triangles

550 triangles

455 triangles

1197 triangles

1969 triangles

2064 triangles

Metadata

B52 plane

Level 1 Level 2 Level 3 Level 4 Fig. 4. Instances of hybrid trancoding model with its metadata and some levels

4.2 The Adaptive Transmission Mechanism

In order to avoid data redundancy between the similar models of a 3Dscene, we treat the similar models, which are identical except with different position, orientation, scale, and material properties, as one element. In the opinion of mesh compression field, the connectivity and geometry is the main shape description, and the others are appearance attributes: texture, color, and etc.. At the server, we use hierarchical element tree to organize the scene (Fig. 5), just like the common modeling tools, i.e.,

A QoS Controller for Adaptive Streaming of 3D Triangular Scenes

579

3D Max, Open Inventor. However, we take the idea of element graph [7], the nodes of an element do not directly contain the data (except the group node), but the information is stored in several pools and the nodes are only reference the entries of the pools. The main difference, between our hierarchical element tree with the element graph [7], is that the structure and function of the pools, since it is more reasonable to organize the pools according to the standard compression classification. 3D Scenes Element E1

Element En Group Node

Group Node

Texture Coordinate Node

Vertex Array Node

Indexed Triangle Node

Geometry Pool

Topology Pool

Y

Z

Z

Level 1

X

Y

Basic Mesh

X

I0

I1

Indexed Texture-coord Node

Texture Node Texture Pool *.* images

I2

Material Node

Material Pool ambient color diffuse color

I0

I1

I2

specular color

Transform Node Transform Pool

emissive color

Position Orientation Scale

Sub-group Node

Fig. 5. An example of a hierarchical element tree, which can have sub-group node. The Vertex Array Node and Texture Coordinate Node reference the geometrical information, the Indexed Triangle Node and Indexed Texture-coord Node reference the topological field, the transform node contains a set of transformation matrixes for similar models, and the others (excluding the Sub-group Node) are appearance attributes.

By far, to each requesting client, the server can transmit the metadata and optional levels of detail. Such method has the advantage that it can provide the geometric complexity of the data being transmitted in the optimal streaming way. However, determining whether a level of a model should be delivered to the client is not a trivial task and is typically left to the server’s QoS controller. According to the 3D QoS controller, the server selects the elements, which have to be transmitted to a specific client. If an element already exists on the client, then it is ignored. Each selected element gets a priority, which is calculated based on the QoS model. As mentioned before, the pools contain the complete visual information of the 3D model. With the help of the selected elements’ “hierarchical element tree”, the server is able to identify the pool entries, which are shared by multiple client users. At first, the primary data of the elements, i.e., the ID and metadata, are transmitted to the appropriate clients. And then, the optimal progressive levels in the nodes are sent and correspondingly marked. In the meantime, each client creates a similar “hierarchical

580

Z. Cheng and S. Jin

element tree” structure, and identify the pool entries that are expected progressive data, with the help of the 3D QoS controller. Consequently, the selection and progressive transmission of the information according to user’s interests are realized in the improved APRR framework, guided by the 3D QoS Controller.

5 Results and Discussion The prototype of the 3D QoS framework has been designed and implemented in a local area network with 10Mbps bandwidth, based on the old APRR architecture[3]. The server is running on one personal computer with 3.0GHz CPU, 2GB DDR memory, and 256MB graphics adaptor, while the four clients PCs have lower hardware configuration with 2.4 GHz Intel Pentium CPU. To guarantee lossless delivery of data the packets will be delivered in the same order in which they were sent, we adopt the prototype based on TCP protocol. To evaluate our implementation and compare with the old APRR prototype, we test the rendering performance of the same synthetic battlefield environment, including a number of military weapon 3D models and a 10x10km terrain area with some building constructions. In the typical application, for the four clients, we compare the two type parameters: startup time and average bitrate in the startup process. The startup time counts the latency time that one client terminal reconstructs and renders the primary scene by the metadata and the progressive levels of the streaming models, and the average receiving bitrate is evaluated by the average transmitting bitrate in the startup process. The table 1 shows the corresponding values of the parameters, and it’s easy to observe that the startup times in this test case are reduced about 20% under the mostly similar average transmitting bitrates. Table 1. Performance comparison Original APRR Client1 Client2 Client3 Client4

startup time (sec.) average bitrate (kbps) startup time (sec.) average bitrate (kbps) startup time (sec.) average bitrate (kbps) startup time (sec.) average bitrate (kbps)

120 452 148 398 85 432 198 535

APRR with QoS 89 466 124 412 71 445 161 517

To test the effect of trading off quality for frame rate on the visual quality of the whole scene, we kept the frame rate fixed and took the snapshot of the scene at several viewpoints in the client 1 (Fig. 6.a and 6.b) and client 2 (Fig. 6.c and 6.d). We then compared the quality of the scene with and without adaptive frame rate. Figure 6 shows the snapshots of the scene at two different viewpoints when the scene is rendered at the maximum quality (Fig. 6.a and 6.c) and when it is rendered at a fixed target frame rate of 30 fps (Fig. 6.b and 6.d). In Fig. 6.b and 6.d, we can see that

A QoS Controller for Adaptive Streaming of 3D Triangular Scenes

(a) 13fps

(c) 17fps

581

(b) 30fps

(d) 30fps

Fig. 6. The quality and frame rate trade-off

degradation is easily observed at the red-ellipse marked areas. Obviously, by applying an almost un-evident quality degradation, we obtain a fixed interactive frame rate. This fact proves the appropriateness of our QoS controller.

6 Conclusion We have presented a QoS framework that enables constant frame rate in the client terminals. In this paper, we have made two main contributions. By analyzing the performance of the remote rendering pipeline for 3D models, we have proposed a QoS architecture for 3D QoS and identified clearly the QoS specification and management of the overall QoS framework. In addition, we have proposed a fast hybrid transcoding algorithm for adaptive streaming of a 3D triangular object, which progressively partitions the object into parts: a series of refiner meshes, and its metadata that is its compressed basic mesh with texture. And the algorithm has been applied in the 3D QoS framework to create the hierarchical element tree of the scene at server, guided by the 3D QoS Controller. One prototype, about the virtual military battlefield environment, has been designed and implemented in the networked PC platforms. The experimental results have proved the soundness of our framework and algorithm guided by the 3D QoS controller. Most of the ideas in this article are emerged during the improvement of the original APRR [3] system. I have presented them in this article with the goals of emphasizing the advantages of adaptive delivery of 3D models and pointing out some of the challenges involved. Although our experiments have led us to solutions to some of the problems, many issues remain open, i.e., out-of-core streaming of large-scale

582

Z. Cheng and S. Jin

environments. With increasing bandwidth, some of the issues in this article may be alleviated but not entirely solved. At any given moment, more information is generated than any computer can handle. Intelligent transcoders will be necessary to exchange it over networks, visualize it, and ultimately, make sense of it. Acknowledgments. We are grateful to Isenberg Martin for the provision of demonstration code of their related paper. This work is supported by National Nature and Science Fund of Hunan province in china (No. 05JJ30124).

References 1. C. O'Sullivan, S. Howlett, R. McDonnell, Y. Morvan, and O. Keith. Perceptually Adaptive Graphics. In Proceedings of Eurographics STAR, 2004. 2. I. B. Martin, and T. J. Watson. Adaptive Graphics. IEEE Computer Graphics and Applications, 2(1): 6-10, 2003. 3. Z.Q. Cheng, S.Y. Jin, G. Dang, T. Yang, and T. Wu. A service-oriented architecture for progressive delivery and adaptive rendering of 3D content. In Proceedings of VSMM， LNCS4270， Xi’an, China, pages 110-118, 2006. 4. G. Hesina, and D. Schmalstieg. A Network Architecture for Remote Rendering. Technical report, TR-186-2-98-02, Institute of Computer Graphics and Algorithms, Visualization and Animation Group, Vienna University of Technology, 1998. 5. N. P. Ngoc, W. van Raemdonck, G. Lafruit, G. Deconinck, and R. Lauwereins. A QoS Framework for interactive 3D Applications. In Proceedings of the 10th International Conference on Computer Graphics and Visualization, pages 317-324, 2002. 6. I. M. Martin. Hybrid Transcoding for Adaptive Transmission of 3D Content. In Proceedings of IEEE Int’l Conf. Multimedia and Expo, pages 373-376, 2002. 7. J. Sahm, Ingo Soetebier, and H. Birthelmer. Efficient Representation and Streaming of 3Dscenes. Computers & Graphics, 28(1):15-24, 2004. 8. M Guthe, and R. Klein. Streaming HLODs:an out-of-core viewer for network visualization of huge polygon models. Computers & Graphics ,28(1):43-50, 2004. 9. M. Isenburg, Y. Liu, J. Shewchuk, and Jack Snoeyink. Streaming computation of Delaunay triangulations. ACM Transactions on Graphics, 25(3):1049-1056, 2006. 10. S. Olbrich, and H. Pralle. High-Performance Online Presentation of Complex 3D Scenes. In Proceedings of the IFIP TC-6 Eighth International Conference on High Performance Networking, pages 471-484, 1998. 11. A. Guéziec, G. Taubin, B. Horn, F. Lazarus. A Framework for Streaming Geometry in VRML. IEEE Transactions on Computer Graphics and Application, 19(2):68-78, 1999. 12. B.-Y. Chen, and T. Nishita. The Development of 3D Graphics and VRML Libraries for Web3D Platform by Using Java. Systems and Computers in Japan, 34(10):47-55, 2003. 13. M. Isenburg, and J. Snoeyink. Binary Compression Rates for ASCII Formats. In Proceedings of Web3D, pages 173-178, 2003. 14. E. Fogel, D. Cohen-Or, R. Ironi, and T. Zvi. A Web Architecture for progressive delivery of 3D content. In Proceedings of Web3D, Monterey, California, USA, pages 35-41, 2004. 15. J. Jessl, M. Bertram, and Hans Hagen. Web-based progressive geometry transmission using subdivision-surface wavelets. In Proceedings of Web3D, pages 29-35, 2005. 16. M． Deering. Geometry compression. In Proceedings of SIGGRAPH, Los Angeles, CA , USA, pages 13-20, 1995.

A QoS Controller for Adaptive Streaming of 3D Triangular Scenes

583

17. H. Hoppe. Progressive Meshes. In Proceedings of SIGGRAPH, New Orleans, LA, USA, pages 99-108, 1996. 18. P. Alliez, and M. Desbrun. Progressive encoding for lossless transmission of triangle meshes. In Proceedings of SIGGRAPH, Los Angeles, CA, USA, pages 198-205, 2001. 19. P. Sander, J. Snyder, S. Gortler, and H. Hoppe. Texture Mapping Progressive Meshes. In Proceedings of SIGGRAPH, Los Angeles, CA, USA, pages 409-416, 2001. 20. I. Guskov, K. Vidimce, W. Sweldens, and P. Schroeder. Normal meshes. In Proceedings of SIGGRAPH, New Orleans, USA, pages 95-102, 2000. 21. P. Alliez, and C. Gotsman. Recent Advances in Compression of 3D Meshes. Advances in Multiresolution for Geometric Modelling, 3-26, 2005. 22. J. Peng, C.-S. Kim, and C.C.J. Kuo. Technologies for 3D Mesh Compression: A Survey. Journal of Visual Communication and Image Representation, 16(6): 688-733, 2005. 23. G. Al-Regib, and Y. Altunbasak. 3TP: 3-D Models Transport Protocol. In Proceedings of Web3D, Monterey, California, USA, pages 155-163, 2004. 24. G. Al-Regib, Y. Altunbasak, and J. Rossignac. Error-Resilient Transmission of 3D Models. ACM Transactions on Graphics, 24(2):182-208, 2005. 25. A. Lall, V. Sekar, M. Ogihara, X. U. Jun, and H. Zhang. Data streaming algorithms for estimating entropy of network traffic. In Proceedings of the joint international conference on Measurement and modeling of computer systems, pages 145-156, 2006. 26. D. Koller, M. Turitzin, M. Tarini, G. Croccia, P. Cignoni, and R. Scopigno. Protected Interactive 3D Graphics Via Remote Rendering. In Proceedings of SIGGRAPH, pages 695-703, 2004. 27. S. Gumhold. Truly selective polygonal mesh hierarchies with error control. Computer Aided and Geometry Design, 22(5):424-443, 2005. 28. S. Rusinkiewicz, and M. Levoy. Streaming QSplat: A Viewer for Networked Visualization of Large, Dense Models. In Proceedings of Symposium on Interactive 3D Graphics. Monterey, CA, USA, pages 61-68, 2003. 29. A. Kalaiah, and A. Varshney. Statistical geometry representation for efficient transmission and rendering. ACM Transactions on Graphics, 24(2):348-373, 2005. 30. J. D. Foley, A. van Dam, S. K. Feiner, and J. F. Hughes. Computer Graphics. Second Edition in C, Pearson Education, USA, 1996. 31. Z.Q. Cheng, S.Y. Jin, and H.F. Liu. Anchors-based lossless compression of progressive triangle meshes. In Proceedings of Pacific Graphics, Taibei, Taiwan, 45-50, 2006. (short paper) 32. M. Isenburg, and J. Snoeyink, Compressing texture coordinates with selective linear predictions. In Proceedings of computer graphics International, pages 126-131, 2003. 33. M. Soucy, G. Godin, and M. Rioux. A texture-mapping approach for the compression of colored 3D triangulations. Visual Computer, 12(10): 503-514, 1986.