Content-Aware Adaptive Streaming
From NMSL
We are designing adaptive streaming algorithms that are based on the visual content of the video streams. The goal is to adaptively transmit the most important frames to clients to yield the best quality. Several ideas are being explored, including: real time and offline processing of video streams, summarization of sports videos, and adaptation of multi-layer scalable video streams.
People
- Mohamed Hefeeda (Assistant Professor)
- Wael Abd-Almageed (Assistant Research Scientist at UMD)
- Majid Bagheri (PhD Student)
Issues
- How to set parameters to detect shot boundaries?
- Different values for parameters result in different peaks in frame significance plots, which setting should be used?
- Which distance metric to be used for computing distortion?
- pixel-based metrics such as MSE, SSIM are too sensitive to camera motions, they do not 'understand' the content
- How many key frames for each shot? based on shot length? motion? significance variation?
- Knowing the target number of key frames how to distribute it among shots?
Discussion and Ideas
- discussion_17mar08.pdf
- Using only HS histogram could help reduce the dimension.
- Explore video quality measures
References and Links
- K. Tan, R. Ribier, S. Liou, Content-sensitive video streaming over low bitrate and lossy wireless network, ACM Multimedia 2001.
- X. Zhou, S. Liou, Optimal nonlinear sampling for video streaming at low bit rates, IEEE Transactions on Circuits and Systems for Video Technology, Jun 2002, 12(6), pp. 535-544.
