Document not found! Please try again

a quality-driven decision engine for live video ... - CiteSeerX

1 downloads 0 Views 1MB Size Report
system modeling and analysis, and approximate dynamic programming. SONG CI [S'98, M'02, SM'06] ([email protected]) is an assistant professor of computer and ...
CI LAYOUT

8/6/09

12:05 PM

Page 48

S E RV I C E -O R I E N T E D B R O A D B A N D WIRELESS NETWORK ARCHITECTURE

A QUALITY-DRIVEN DECISION ENGINE FOR LIVE VIDEO TRANSMISSION UNDER SERVICE-ORIENTED ARCHITECTURE DALEI WU, SONG CI, AND HAIYAN LUO, UNIVERSITY OF NEBRASKA-LINCOLN HAOHONG WANG, MARVELL SEMICONDUCTORS AGGELOS KATSAGGELOS, NORTHWESTERN UNIVERSITY

ABSTRACT

(a)

The authors propose a service-oriented decision engine framework, which consists of a decision engine, a performance evaluation component, and other major SOA components to support real-time video transmission over wireless multi-hop networks. 48

Service-oriented architecture provides a solution to the increasing network complexity due to ever-growing heterogeneous networks. As the most significant component of SOA, the decision engine is to create a workflow, defined as a sequence of individual data processing entities, for providing end-to-end QoS of a given task. Although the workflow of video transmission is generally known, existing solutions are often monolithic. Furthermore, there is no decision engine to select a workflow based on the best user-perceived quality. In this article we propose a service-oriented decision engine framework, which consists of a decision engine, a performance evaluation component, and other major SOA components to support real-time video transmission over wireless multihop networks, aiming to provide the best user-perceived video quality under application-centric QoS constraints. Based on the investigation of the stateof-the-art research efforts on SOA, some key issues for wireless live video transmission are discussed, and a case study for live video transmission is given to illustrate the proposed scheme. The superior performance of the proposed service-oriented decision engine is validated by experimental results.

INTRODUCTION In the past few years, broadband wireless networks have been gaining popularity due to their features of high data rate and large coverage. The demand for higher data rates stems mainly from the need to stream multifarious high-quality multimedia content to mobile users. However, video transmission in both wired and wireless networks still largely adopts the traditional monolithic design methodology, meaning that video data processing and delivery are done at the source. This makes it very difficult to sup-

1536-1284/09/$25.00 © 2009 IEEE

port emerging video applications on a large scale such as videoconferencing and live video broadcasting/multicasting, especially under a dynamic heterogeneous network environment. Furthermore, compared to traditional data communications such as data file transfer and download, video communications has special characteristics. For example, different bits in a video stream have different levels of importance in terms of their contribution to user-perceived video quality. In addition, the error resilience and error concealment (EC) techniques used in the process of video encoding and decoding allow multimedia applications to recover from a certain degree of packet losses. These unique features make network-centric metrics, such as throughput, average packet loss rate, and average packet delay, unsuitable to represent and evaluate the quality of service (QoS) of multimedia communications. Moreover, for many existing designs in multimedia networking, an end user may not be able to obtain the desired QoS because most multimedia bearer services are provided in a best effort way. In recent years service-oriented architecture (SOA) has been widely regarded as a promising distributed network management method in large-scale heterogeneous communications networks [1, 2]. From the standpoint of SOA, the entire video communication system can be decomposed into many different services provided by one or more service providers. In other words, platforms, software, and data are all treated as services. To provide the required application-centric QoS, a chain of services (workflow) has to be created by a decision engine, shown in Fig. 1, which will utilize information collected by other SOA components such as service discovery, session control, service validation, and performance evaluation, as well as a global brokerage platform. Therefore, the desire to provide high-quality service and a better end-user experience calls for a service-

IEEE Wireless Communications • August 2009

Authorized licensed use limited to: Northwestern University. Downloaded on December 16, 2009 at 13:31 from IEEE Xplore. Restrictions apply.

CI LAYOUT

8/6/09

12:05 PM

Page 49

oriented quality-driven framework to direct the design and evaluate the performance of multimedia communications systems in heterogeneous networks. In general, video applications can mainly be classified into two categories. One category is video streaming applications, such as iTunes [3] and YouTube [4], where the media server resides at a different location from router/gateway, and the video is pre-encoded and packetized at the same server. Therefore, video encoding cannot be adapted to changes in the network such as network congestions and/or channel variations. Another category is interactive video applications, such as videoconferencing, video instant messaging (e.g., Skype and MSN), and live broadcasting, where the captured videos are coded on-the-fly, and the source content and network conditions are jointly considered to determine the optimal source encoding modes [5]. More importantly, online video coding allows us to allocate more resources to the video contents that have more impact on user-perceived video quality, while spending fewer resources on those that are less important. Therefore, from the SOA point of view, different classes of applications require different workflows to provide the required end-to-end QoS [2]. Multimedia communications consist of continuous data having temporal and spatial data dependencies. Thus, the monolithic implementation of most existing multimedia systems makes it very difficult to build a large-scale multimedia system, usually incurring much costly and timeconsuming work [2]. In addition, existing SOA for multimedia communications lacks real-time capabilities. Specifically: • It is not dynamically configurable and adaptable to meet real-time requirements. • Timing and interaction issues are not thoroughly studied and formally expressed in existing SOA for multimedia communications. • Web services lack real-time capabilities. • Network awareness and control login are not integrated into the application service login. In this work we provide a quality-driven decision engine of service-oriented architecture for delay-bounded multimedia delivery. We focus on real-time video applications and explore an SOA framework aimed at optimizing video transmissions in wireless multihop networks. The proposed SOA consists of the key data processing components of multimedia signal processing, communication network, and performance evaluation. The multimedia signal processing component provides a variety of signal processing services, such as video transcoding and content analysis. The communication network provides bearer services of media delivery, including transmission path selection, network resource allocation and scheduling, and network QoS provisioning. The performance evaluation component can predict and evaluate the current user-experienced video quality. Based on the expected user-perceived video quality, the proposed decision engine will combine the services provided by one or more components optimally into a workflow to provide the best video service to an end user. In this work we use content-aware video transmis-

Media server 2

Media server 1

Media signal processing

UDDI Decision engine

Global broker network

Service discovery

Session control

Performance evaluation

Workflows

Validation engine

Execution engine

User devices

 Figure 1. The proposed service-oriented architecture for live video communications.

sion as a case study to illustrate the concept of the proposed decision engine under the SOA framework. The rest of the article is organized as follows. We first describe the system model of the proposed quality-driven SOA framework for realtime video communications in multihop wireless networks. Then the different services provided by each system functional entity are described in detail. To illustrate how the proposed decision engine optimally combines these services into a quality-aware workflow, a case study for the proposed SOA-based live video transmission framework is discussed. Then the performance of the proposed framework is evaluated by experiments before the conclusion.

PROPOSED SOA SYSTEM FOR LIVE VIDEO TRANSMISSION In this section we describe the proposed SOA system model and the different services provided by each component of the SOA system.

THE PROPOSED SOA SYSTEM MODEL The system model of the proposed service-oriented architecture for live video transmission is shown in Fig. 2 where the decision engine communicates with the media signal processing services and the communication network services through broker network management. The performance evaluation service evaluates the expected user-perceived video quality. Based on the evaluated expected video quality, the decision engine can therefore combine the available services to compose a workflow to achieve the best user-perceived video quality. Specifically, the decision engine can retrieve the user profile information and services from the broker network, optimize the service configuration, and implement different capacities of applications. For example, end users can connect to the net-

IEEE Wireless Communications • August 2009 Authorized licensed use limited to: Northwestern University. Downloaded on December 16, 2009 at 13:31 from IEEE Xplore. Restrictions apply.

49

CI LAYOUT

8/6/09

12:05 PM

Page 50

Resource allocation Scheduling

Performance evaluation service

Decision engine

Workflows

Transmission path selection

Communication network services

Network QoS provisioning

Broker network management

Media signal processing service

 Figure 2. The proposed SOA system model for live video communications, where the decision engine combines the available services from the broker network into optimal workflows based on the information from the performance evaluation component.

work through the Internet, WiFi, code-division multiple access (CDMA), or Universal Mobile Telecommunications System (UMTS). As long as the user profile information is known, the decision engine can retrieve all the available services from the service pool that media signal processing providers and networks provide under the resource constraints of the current network scenario, and provide end users with the best possible video quality while satisfying the underlying resource limits. The major services provided by the proposed architecture, shown in Fig. 2, are discussed in the following sections.

MEDIA SIGNAL PROCESSING SERVICES Based on different user profiles and available network resources, the decision engine selects different media signal processing algorithms (services) to deal with user requests from the media servers. Some examples of video processing operations that could be offered by a thirdparty video processing service provider are: • Extracting the regions of interest (ROI) information of a video sequence • Downsampling a frame • Filtering the high-frequency component of a frame • Encoding or transcoding a video sequence • Summarizing a video sequence • Dropping the current frame

PERFORMANCE EVALUATION SERVICE From the point of view of end users, applicationcentric metrics such as the expected end-to-end video quality are the most straightforward and reasonable utility function of any optimization framework for video communications [6]. Traditional network-centric metrics such as network throughput, average packet delay, and average packet loss rate fail to provide an efficient and accurate evaluation on the performance of video transmissions. The reason is that network-centric metrics ignore the intrinsic requirements of live

50

video communication. These features and requirements are: • The packetization scheme at the source has a significant impact on the user-perceived video quality. • Not all the bits of a coded video bitstream are of equal importance in determining the userperceived video quality. • Enabling continuous and smooth playback is very important to achieve a good user experience. • There are various available error resilience and error concealment techniques to enhance received video quality under poor network conditions. Therefore, an application-centric performance evaluation method is necessary to provide such a service for the decision engine to configure other services and form optimal workflows for end users.

NETWORK SERVICES Path Selection — There are multiple paths in a heterogeneous multihop network that may provide different levels of transmission reliability [7]. Therefore, one kind of service is to make intelligent routing choices and offer reliable data delivery. In heterogeneous network scenarios the decision engine will integrate some existing routing protocol, such as optimal link state routing (OLSR), into a workflow of an end-to-end session to find the optimal transmission path. Therefore, the decision engine is to select the best routing service to determine a transmission path based on the utility function provided by the performance evaluation component. Resource Allocation and Scheduling — Multimedia data of a given video stream have different levels of importance in terms of their contribution to the user-perceived video quality. Since in most cases network resources are limited, the necessary service provided by a network is to make sure that important multimedia data are allocated with more network resources than less important multimedia data. In heterogeneous scenarios various resource allocation and scheduling approaches have been developed and implemented in different types of networks, such as time slot and/or bandwidth allocation, packet ordering, and retransmission. The decision engine needs to choose an approach such that the user-perceived video quality is maximized while the utilization of network resources is enhanced.

CASE STUDY: AN SOA-BASED LIVE VIDEO COMMUNICATION SYSTEM We consider an N-frame video sequence C = {g 1 , …, g N }. We assume that the media signal processing provider can perform content analysis; that is, the content of each video frame can be divided into a foreground part and a background part with the foreground part being the ROI. Furthermore, the foreground and background parts can be encoded with different coding options.

IEEE Wireless Communications • August 2009 Authorized licensed use limited to: Northwestern University. Downloaded on December 16, 2009 at 13:31 from IEEE Xplore. Restrictions apply.

CI LAYOUT

8/6/09

12:05 PM

Page 51

In this work, without loss of generality, we consider a multihop wireless network modeled as a directed acyclic graph (DAG) G(V, E) with node set V and edge set E. To provide a smooth video display experience to end users, the delay deadline imposed on the transmission of each packet k over G is associated with frame decoding deadline Tmax. In live video communications, frame decoding deadline T max is linked with frame rate r as T max ≈ 1/r, where frame rate r comes from the QoS requirement r in the end user’s profile. The frame decoding deadline Tmax indicates that all the packets needed to decode a frame must be received at the decoder buffer prior to the playback time of that frame. We assume that Tmax is known to the decision engine by message exchange to perform optimization. In real transmission, when packet k reaches an intermediate node v, the network always checks the total delay t kv of packet k at node v. If t kv exceeds T max, packet k should not get through node v and should be discarded. In this work we propose that the performance evaluation uses the pixel recursive algorithm (ROPE) [8] to estimate the expected end-to-end distortion at the receiver. The accuracy of ROPE in end-to-end distortion estimation is attributed to its ability to recursively calculate the first and second moments of the decoder reconstructed pixels. In addition, since foreground and background packets have different contributions to the user’s playback experience, the contributions of their distortion to the user-perceived video can be weighted by λk by the performance evaluation service. From the user’s point of view, λk of foreground packets is much bigger than that of background packets. Considering different impacts of foreground and background packets on user-perceived video quality, when both foreground and background packets are queued at intermediate node v, the network follows a priority scheduling approach in which the foreground packet is first scheduled for transmission by the network. Therefore, the scheduling service Φk for packet k is provided by the network and determined by the decision engine, based on the video quality evaluation result of the performance evaluation service. In addition, each link is assumed to be able to perform packet-based retransmission. The maximum number of retransmissions Π(v,u) for packet k over k link (v,u) is jointly determined by the packet delay constraint Tmax and the total delay tkv packet k has experienced before it reaches the head of the queue at node v. The packets whose delay constraints have been violated are dropped from the queue. Finally, each packet k generated by the media signal processing service and transmitted by the network is characterized by: • The source coding service Sk • The transmission path selection service Pk • The scheduling service Φk • The packet delay deadline Tmax • The quality impact factor λk Thus, the expected received video distortion E[D k ] for packet k can be written as E[D k ] = Q k ( S k , P k , Φ k , T max , λ k ) by the performance evaluation service. Based on different available services, the decision engine needs to optimize the combination of these services for an application session. We assume that video clip C

(a)

(b)

(c)

(d)

 Figure 3. The identification of ROI. requested by an end user is compressed into the b b packet group {k 1f, … , k If, k I+1 , … , k I+J }, which comprises I foreground packets and J background packets. Therefore, the goal of the decision engine is to jointly determine the optimal services, including the optimal coding service S*k, the optimal transmission path selection service P k*, and the optimal scheduling service Φ k* for each packet k of the whole video clip to maximize the expected user-perceived video quality under a delay constraint. For notation simplicity, we define the following optimal service vector for each packet k: V*k = {S*k, Pk*, Φk*}. Therefore, the goal of the decision engine can be formulated to find

{

}

V := Vk* k = 1… I + J = arg min

{

}

I+J



k =1

E ⎡⎣ Dk ⎤⎦

s.t .: max E ⎡⎣t1 ⎤⎦ ,, E ⎡⎣t1+ J ⎤⎦ ≤ T max

(1)

where tk is the end-to-end delay of packet k, and V is the generated workflow by the decision engine for the end user of video sequence C. Most decoder concealment strategies introduce dependencies among packets in a video session. To solve the constrained multistage optimization problem (1), we propose using Lagrangian relaxation (LR) and dynamic programming (DP) techniques to solve the formulated problem in the decision engine [9].

EXPERIMENTAL RESULTS In this section experiments are designed using H.264/AVC JM 12.2 [10] for the video clip called “Mother and Daughter.” The proposed decision engine framework under SOA is evaluated in a

IEEE Wireless Communications • August 2009 Authorized licensed use limited to: Northwestern University. Downloaded on December 16, 2009 at 13:31 from IEEE Xplore. Restrictions apply.

51

CI LAYOUT

8/6/09

12:05 PM

Page 52

30-node network deployed over a 1000 m × 1000 m rectangular region. The source node and destination node are chosen randomly from the nodes in the network. The connectivity between the nodes is determined by the radio transmission range. The transmission range for each node is assumed to be 150 m. We generate 50 topologies and run 50 computations to obtain the average results. Before video encoding, the identification of the ROI is performed by the following stages during the experiments: background subtraction, split-and-merge classification, and morphological operations. The results of the different stages of the above operation are shown in Fig. 3. Based on the content-aware analysis result, the network prioritizes the fore38

Quality-driven PLR-based Delay-based

36

34

PSNR (dB)

32

30

28

26

24 22 5

10

15

20

25

SINR (dB)

 Figure 4. PSNR vs. SINR for different routing algorithms.

Without content analysis service With content analysis service — foreground With content analysis service — background

40

PSNR (dB)

35

30

25

20

0

With priority scheduling

Without priority scheduling

 Figure 5. PSNR performance comparison between the schemes with and without using priority scheduling.

52

ground parts by allocating more network resources to foreground packets against the less important background parts under a resourcelimited wireless environment. Therefore, the proposed service-oriented live video transmission system manages to improve the user-perceived video quality even in a resource- limited network. In the experimental analysis, the peak signal-to-noise ratio (PSNR) is used to measure the user-perceived video quality at the receiver node. First, we compare the impacts on user-perceived video quality of different routing services provided by the network. The quality-driven (application-centric) routing service using the expected end-to-end distortion as the routing metric is compared with the network-centric routing services. Here, two types of network-centric routing services are considered, one using the link packet loss rate as the routing metric (PLR-based) and the other using the link packet delay as the routing metric (delay-based). To guarantee fair comparison, video coding optimization is also performed with both the PLRbased and delay-based routing services. The PSNR of the received video is calculated for each frame and is averaged over all frames. Figure 4 shows the PSNR comparison with different average link signal-to-interference-plus-noise ratio (SINR) and packet delay deadline Tmax = 0.033 s. The results clearly show that the qualitydriven routing service achieves significant performance gains over the two network-centric routing services. This is because quality-driven routing service aims at minimizing the video distortion in determining the optimal path, while minimizing the packet loss rate in the PLRbased routing service or minimizing the packet delay in the delay-based routing service does not always lead to minimized video distortion. Recall that the foreground packets are first scheduled when both the foreground and background packets are simultaneously present in the same queue of a network node. To compare the received quality of different video contents, two schemes are considered. The first scheme uses the content analysis service to perform the identification of ROI (IRI), while the second scheme does not use the content analysis service, considering the requested video sequence as a whole part under no IRI. Figure 5 shows the PSNR performance of the above two schemes with and without using the service of priority scheduling. The single-packet delay deadline Tmax is 0.01 s. As expected, without using the service of priority scheduling, the average PSNRs of the whole video under no IRI, the foreground, and the background are the same. However, when the service of priority scheduling is used, the foreground has a 4.5 dB and 9.5 dB PSNR improvement over the whole video under no IRI and the background, respectively. This indicates that the proposed decision engine truly provides a guarantee for the received quality of ROI in a delaystringent or rate-limited multihop wireless network. This can also be verified by a statistical result in our experiment, which shows that the number of lost foreground packets is only 51 percent that of lost background packets, while the average packet delay of foreground packets

IEEE Wireless Communications • August 2009 Authorized licensed use limited to: Northwestern University. Downloaded on December 16, 2009 at 13:31 from IEEE Xplore. Restrictions apply.

CI LAYOUT

8/6/09

12:05 PM

Page 53

(a)

(b)

(c)

 Figure 6. Comparison of reconstructed frames: a) the original; b) using content analysis and priority scheduling; c) without using content analysis and priority scheduling. is consistently smaller than that of background packets. To show the received video quality at the destination, we plot the 61st frame of the “Mother and Daughter” sequence obtained in different scenarios in Fig. 6. In this figure, a is the original video frame, b is the reconstructed video frame with using the services of content analysis and priority scheduling, and c is the reconstructed video frame without using the services of content analysis and priority scheduling. In both b and c, T max is 0.005 s. It is observed that the scheme using the services of content analysis and priority scheduling provides better user-perceived quality than the scheme without using the services of content analysis and priority scheduling.

CONCLUSIONS Traditional multimedia communication systems are monolithic, lacking the flexibility of end-toend QoS provisioning for various multimedia applications, especially for live video applications on a large scale under a heterogeneous network environment. In this work we propose a quality-driven decision engine for real-time video transmissions based on SOA, where various kinds of data processing services are jointly considered and optimized by the proposed decision engine. The proposed design is illustrated by the case study of content-aware quality-driven workflow generation. Depending on the different contributions to the user-perceived video quality, regions of interest (foregrounds) of video frames are extracted from the background. With the expected end-to-end user-perceived quality as the objective function, various services such as video encoding, transmission path selection, and packet scheduling for foregrounds/ backgrounds are put together in a systematic way. Experimental results show that the proposed quality-driven service-oriented decision engine can provide a much better end-user experience than existing approaches.

ACKNOWLEDGMENTS This material was partially supported by NSF under Grant CCF no. 0830493 and CRI no. 0707944. It was also partially supported by Layman Foundation.

REFERENCES [1] W. Wu et al., “Service Oriented Architecture for VoIP Conferencing,” Special Issue on Voice over IP — Theory and Practice, Int’l. J. Commun. Sys., vol. 19, no. 4, May 2006, pp.445–61. [2] I. Brunkhorst, S. Tonnies, and W. Balke, “Multimedia Content Provisioning using Service Oriented Architectures,” Proc. IEEE ICWS, Beijing, China, Oct. 2008. [3] iTunes; http://www.itunes.com/ [4] YouTube: http://www.youtube.com/ [5] A. Katsaggelos et al., “Advances in Efficient Resource Allocation for Packet-Based Real-Time Video Transmission,” Proc. IEEE, vol. 93, no. 1, Jan. 2005, pp.135–47. [6] Y. Andreopoulos, N. Mastronade, and M. van der Schaar, “Cross-Layer Optimized Video Streaming over Wireless Multihop Mesh Networks,” IEEE JSAC, vol. 24, no. 11, Nov. 2006, pp.2104–15. [7] S. Mao et al., “Multipath Routing for Multiple Description Video in Wireless Ad Hoc Networks,” Proc. IEEE INFOCOM, Miami, FL, Mar. 2005. [8] R. Zhang, S. L. Regunathan, and K. Rose, “Video Coding with Optimal Inter/Intra-Mode Switching for Packet Loss Resilience,” IEEE JSAC, vol. 18, no. 6, June 2000, pp. 966–76. [9] D. Wu, S. Ci, and H. Wang, “Cross-Layer Optimization for Video Summary Transmission over Wireless Networks,” IEEE JSAC, vol. 25, no. 4, May 2007, pp. 841–50. [10] JM; http://iphome.hhi.de/suehring/tml/

BIOGRAPHIES D ALEI W U [S‘05] ([email protected]) received his B.S. and M.Eng. degrees in electrical engineering from Shandong University, Jinan, China, in 2001 and 2004, respectively. He is currently working toward a Ph.D. degree in the Department of Computer and Electronics Engineering, the University of Nebraska-Lincoln. His research interests include cross-layer design and optimization over wireless networks, wireless multimedia communications, large-scale system modeling and analysis, and approximate dynamic programming. SONG CI [S’98, M’02, SM‘06] ([email protected]) is an assistant professor of computer and electronics engineering at the University of Nebraska-Lincoln. He received his B.S. from Shandong University of Technology (now Shandong University), Jinan, China, in 1992, his M.S. from the Chinese Academy of Sciences, Beijing, in 1998, and his Ph.D. from the University of Nebraska-Lincoln in 2002, all in electrical engineering. He is director of the Intelligent Ubiquitous Computing Laboratory (iUbiComp Lab) and holds a courtesy appointment with the UNL Ph.D. in Biomedical Engineering Program. He is also affiliated with the Nebraska Biomechanics Core Facility at the University of Nebraska at Omaha and the Center for Advanced Surgical Technology (CAST) at the University of Nebraska Medical Center, Omaha. His research interests include dynamic complex system modeling and optimization, green computing and power management, dynamically reconfigurable embedded systems, content-aware quality-driven cross-layer optimized multimedia over wireless, cognitive network management and service-oriented architecture, and cyber-enabled ehealthcare.

IEEE Wireless Communications • August 2009 Authorized licensed use limited to: Northwestern University. Downloaded on December 16, 2009 at 13:31 from IEEE Xplore. Restrictions apply.

53

CI LAYOUT

8/6/09

12:05 PM

Page 54

H AIYAN L UO [S‘09] ([email protected]) is a Ph.D. student with the Computer and Electronics Engineering Department at the University of Nebraska-Lincoln. His research interests focus on distributed systems and networking. He has served as a referee for IEEE Transactions on Wireless Communications, IEEE Transactions on Vehicular Technology, and several conferences such as IEEE GLOBECOM and IEEE ICC. Previously, he worked for Bell Laboratories, Lucent Technologies as a member of technical staff. He holds a master’s degree from Dalian University of Technology, and received his Bachelor’s degree from Dalian Jiaotong University (formerly Dalian Railway Institute). H AOHONG W ANG [S’03, M‘04] ([email protected]) received a B.S. degree in computer science and an M.Eng. degree in the computer and its application, both from Nanjing University, China, an M.S. degree in computer science from the University of New Mexico, Albuquerque, and a Ph.D. degree in electrical and computer engineering from Northwestern University, Evanston, Illinois. He is currently a system architect at Marvell Semiconductors, Santa Clara, California. His research areas are multimedia information processing and communications. He is a member of the IEEE Visual Signal Processing and Communications Techni-

54

cal Committee, IEEE Multimedia and Systems Applications Technical Committee, and IEEE Multimedia Communications Technical Committee. He is the Founding Chair of the Steering Committee of the annual International Symposium on Multimedia over Wireless. AGGELOS K. KATSAGGELOS [F‘98] ([email protected]. edu) is a professor of electrical engineering and computer science at Northwestern University. He received a Diploma degree in electrical and mechanical engineering from the Aristotelian University of Thessaloniki, Greece, in 1979, and M.S. and Ph.D. degrees, both in electrical engineering, from the Georgia Institute of Technology, Atlanta, in 1981 and 1985, respectively. His current research interests include multimedia signal processing (e.g., image and video recovery and compression, audio-visual speech and speaker recognition, indexing, and retrieval), multimedia communications, computer vision, pattern recognition, and DNA signal processing. He has served the IEEE and other professional societies in many capacities. He is currently a member of the IEEE Technical Committee on Visual Signal Processing and Communications, and the Editorial Boards of the Academic Press, Marcel Dekker Signal Processing Series, the International Journal on Image and Video Processing, and Advances in Multimedia.

IEEE Wireless Communications • August 2009 Authorized licensed use limited to: Northwestern University. Downloaded on December 16, 2009 at 13:31 from IEEE Xplore. Restrictions apply.

Suggest Documents