1. Study-Element Based Adaptation of Lecture Videos to Mobile Devices Ganesh Narayana Murthy (M.Tech IIT Bombay) Sridhar Iyer (Associate Professor, IIT Bombay) Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20101

2. Problem Definition Adapt CDEEP videos to be viewable on mobile devices: – Viewable at low network bandwidths (like GPRS) – Viewable at low cost Video bit-rate – Size of video stream over time – Total size = bit-rate * total time – CDEEP video bit-rate: 1150kbps – GPRS bit-rate: 40kbps The problem: – Video playing incurs delays if available network bandwidth is less than video-bit rate Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20102

3. Video Transcoding Converting from one video format to another – Changing video bit rate – Changing other parameters like frame rate, screen resolution Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20103 Format NameTypical Bit RateApplication MPEG-11.5Mbps or lessCD-ROM MPEG-25-8MbpsDVD, HDTV H.263Typically low bit ratesLow bit-rate video conferencing MPEG-4 / H.26440Kbps to 10Mbps and above Internet Streaming, Video Telephony Flash Video (FLV)Typically low bit ratesEmbedded video in websites

4. Video Quality at low-bit rates Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20104 (a) MPEG-1(b) MPEG-2 Images from transcoded videos (Target bit rate : 40kbps, No audio)

5. Video Quality at low bit-rates (contd.) Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20105 (c) H.264 (mp4) (d) H.263 (3gpp) Images from transcoded videos (Target bit rate : 40kbps, No audio)

6. Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20106 (e) Flash Video (flv) Images from transcoded videos (Target bit rate : 40kbps, No audio)

7. Comparison of Video Codecs Format Name Original Video Size Converted Video Size Video Quality at low-bit rates Remarks MPEG-1432MB26MBPoorCannot be used at low bit-rates MPEG-2432MB29.12MBPoorCannot be used at low bit rates H.263432MB38.3MBPoorCannot be used at low bit rates H.264432MB16.9MBGoodProcessing power / Decoding complexity is high.[1] Flash432MB20.5MBGoodCan be used, but cost is still high. Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20107 (Note: Video bit rate = 1150kbps, No audio, Target bit rate = 40kbps, No audio) Video Sizes are still high for viewing over GPRS

8. Study-Element Based Adaptation Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20108

9. Motivation CDEEP video usually consists of – Presentation slides – Instructor explaining on white paper – Video of instructor talking Presentation slide is usually not changing – Video of slide is not required. One image is sufficient Idea – Extract one image every ‘n’ seconds and send to client. – This would reduce amount of data sent for showing one slide. Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/20109

10. Method-1 Send one image every ‘n’ seconds – Server sends one image every ‘n’ seconds to client – Audio is simultaneously streamed Network bandwidth and Size – Network Overhead (NO) = Image Size / n – Size Overhead (SO) = Total size of images What is the user experience? Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201010

11. User Experience Basis Presentation Study Element – Portion of video showing one slide White Paper Study Element – Portion of video showing instructor writing on white paper Instructor Study Element – Portion of video showing instructor talking Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201011 ………….. 051015 Presentation Slide Delay in start of slide 3 ……….. 253035 White Paper Video Time (secs)

12. User Experience Presentation Element – Delay Experienced (D2) = Delay in start of slide as compared to audio White Paper Element – Delay Experienced (D1) = Delay between any two consecutive images = Sending Rate Instructor Element – Only audio important. No image need be sent. User Experience is assumed to be one User Experience (U i ) = 1 sec / D i Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201012

13. Method-2 Trade-off for user experience Cost incurred in terms of number of images sent Same sending interval for all elements, cannot balance user experience and cost. Choose different sending interval for each study element Probably: – Higher user experience for white paper element – Lower user experience for presentation element Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201013 User Experience Cost Sending Interval Trade-Off Relation

14. System Overview Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201014

15. Building the index Corpus of 10 videos – Representative of various departments Consider different sending intervals ‘r’ – For each ‘r’ find NO,SO and U for every study element in a video. – Repeat for all videos and take average. This relation can be used backwards: – For calculating sending interval, given network bandwidth and user experience. Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201015

16. Graphs of User Experience Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201016

17. Graphs of overheads Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201017

18. Results Original Video Size(MB) Images Size (MB) Reduction (%) U1U2Supported Network Bandwidth 4322.8597%0.20.3820kbps and above Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201018 Achieved Size Reduction Fig: Video stream size reduction (note: Original video bit-rate = 1150kbps, No audio)

19. Results (contd.) Sending IntervalU1U2NO1 (kbps ) S01 (MB) NO2 (kbps ) SO2 (MB) Total Size (SO) White Paper Element Presentation Element 550.20.33715.121.2523.431.65.46 5150.20.11515.121.257.810.531.78 15 0.0670.1155.2540.7817.631.031.81 Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201019 Balance User Experience and Cost Required Network Bandwidth =max(NO1,NO2) is reduced Reduction in size user experience for white paper element remaining same

20. Conclusion Large size reduction can be achieved by using the concept of slideshows Identifiying study-elements within the video helps define user-experience of the slideshow. CDEEP Lecture videos can be adapted to low network bandwidths and in a cost-controlled manner. Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201020

21. Future Work Automated tagging – Identifying study element boundaries – Shot detection techniques User Experience Correlation – Identifying relation between obtained user experience and actual user values Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201021

22. References 1. H.264 white paper. http://ati.amd.com/products/pdf/h264_whitepaper.pdf. 2.Real-time Content-Based Adaptive Streaming of Sports Videos. Shih-Fu Chang, Di Zhong, and Raj Kumar. In CBAIVL '01: Proceedings of the IEEE Workshop on Content-based Access of Image and Video Libraries (CBAIVL'01), page 139, Washington, DC, USA, 2001. IEEE Computer Society. 3.Content-aware video adaptation under low-bitrate constraint. Ming-Ho Hsiao, Yi-Wen Chen, Hua-Tsung Chen, Kuan-Hung Chou, and Suh-Yin Lee. EURASIP J. Adv. Signal Process,2007(2):27-27, 2007. 4.A Characteristics-Based Bandwidth Reduction Technique for Pre-recorded Videos. Wallapak Tavanapong and Srikanth Krishnamohan. In IEEE International Conference on Multimedia and Expo (III), pages 1751-1754, 2000. Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201022

23. Questions? Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201023

24. Content-Aware Adaptation Ganesh Narayana MurthyNCC 2010, Chennai, 30/01/201024 Method NameAdaption MechanismVideo QualityRemarks Hsiao et.al.[2]Identify visual attention regions in a frame. Encode them at high quality. PoorQuality of important objects still depends on network bandwidth Chang.et.al [3]Identify events in sports videos at high quality. Other regions as slideshows. GoodSlideshow of images reduces network bandwidth and size Tavanapong. et. Al. [4] Identify non-changing portions of lecture video and extract one image from them GoodExploits redundancy in lecture videos