Sriram Sethuraman

Dynamic frame-rate selection for live LBR video encoders using trial frames

Low bit-rate live encoding and streaming of video over wired or wireless channels, as utilized in place-shifting applications, need to adapt the frame rate and bit-rate according to content and channel conditions. Typical rate control schemes work with a given target frame rate and can result in irregular frame skips that are induced by buffer conditions and quantizer saturation. Prior methods proposed to address the dynamic frame-rate selection (DFS) problem work by projecting the impact of a new frame-rate from the operating frame-rate and can have slower convergence to the required frame-rate. In this paper, we propose a scheme that utilizes a given computational capacity (typical in embedded appliances) to actually assess the impact of the new frame-rate by coding at the corresponding new frame-skip (referred to as trial frames) before committing to it. The proposed scheme offers quick content-adaptive convergence to the target frame-rate and maintains a motion-adaptive frame skipping pattern to provide a subjectively pleasing spatio-temporal quality trade-off at the operating bit-rate. In addition, the scheme can work with most existing rate control methods that operate at a supplied frame-rate and is also suitable for hooking to a user controlled quality dial.

Low-complexity frame-level joint source-channel distortion optimal, adaptive intra refresh

Error resilient, low latency video coding for interactive video applications requires progressive intra coding of macroblocks (MBs) to contain the error propagation. Both refresh MB selection (RMS) and refresh rate selection (RRS) impact the subjective video quality in the presence of packet losses. Joint source-channel rate distortion optimization methods attempt to find the best trade-off between compression efficiency and end-to-end distortion at an MB-level and are typically computationally expensive in addition to not being optimal at a picture level. While probabilistic error propagation tracking is used for refresh MB selection in previous work, these picture-level optimal RRS methods model source-channel distortion by mimicking the effect of periodic intra frame coding which does not match well with content adaptive refresh MB selection. In this paper, we propose a frame-level approach to RRS that aligns the joint source-channel rate-distortion trade-off modeling with an enhanced RMS process to achieve an optimal end-to-end distortion that is content, bit-rate and channel adaptive. For typical videoconferencing content, the proposed approach is quite low in complexity, works on par with off-line multi-pass identification of the optimal fixed refresh rate and is quite competitive when compared to the H.264 joint modelpsilas lossy rate-distortion optimization technique.

Efficient Alternative to Intra Refresh using Reliable Reference Frames

Motion compensated prediction of video results in decoder-side error propagation (EP) in the presence of channel losses. Progressive intra-refresh methods control the EP, but lead to a drop in coding efficiency (CE). Error-resilient video coding schemes using multiple reference frames have been proposed earlier to achieve a better CE. Some schemes (e.g. ACK-based) tend to be conservative at the expense of CE. Methods using probabilistic models and motion tracking to estimate the decoder state offer better CE, but do not take care of the time taken to recover from losses and the EP due to the use of unreliable reference frames. In this paper, we propose an alternative to intra-refresh that controls the EP and overcomes the CE loss through the use of reliable (i.e. near drift-free) reference frames to code the refresh macroblocks. The proposed scheme has been validated through integration into a ITU-T H.264 encoder and has been shown to give a CE gain of up to 1.5dB. Further, it is of low-complexity and suited for implementation on embedded platforms.

Low-cost wireless projector interface device using TI TMS320DM270

Personal computers today have 802.11 (b/a/g) -based wireless connectivity. By connecting laptops to a projector wirelessly, multiple persons could connect to the projector and present the content on their laptops. This typically requires another personal computer at the projector end that is wirelessly connected to the laptop and which receives, decodes, and renders a compressed version of what was rendered on the display of the laptop. On the laptop, the screen buffer is captured and the captured frames are compressed. The encoded stream is streamed over the WLAN to a streaming client. The article presents a low cost interface device for the projector system based on the TI TMS320DM270 image/video processor.

MPEG-2 to H.264 transcoder on TI TMS320DM642

The technology behind H.264, a joint video coding standard by ITU-T and ISO, offers a significant reduction in bit-rate over earlier standards-based technologies such as MPEG-2 and MPEG-4. Such a compression efficiency advantage spurs its adoption in personal video recorders and in network edge servers that stream content over various last mile solutions that have limited bandwidth. This requires transcoding from an MPEG-2 format which is the most prevalent format that is used in terrestrial/satellite/cable transmission of digital television content. The substantial difference between MPEG-2 and H.264 necessitates almost a complete decoding followed by a re-encoding solution, instead of a compressed domain or minimal draft based low-complexity transcoding solution. We present an embodiment of a low-complexity MPEG-2 to H.264 transcoding algorithm and a general-purpose DSP based solution implemented on TIs TMS320DM642 processor