Fast Encoding Techniques For WinMobile VC-1 Encoder (29-Aug-2008)

            In order to achieve high encoding efficiency and high quality for offline encoding or quality targeted encoding, an exhaustive search and complex computations need to be performed to make the best encoding decisions in motion estimation, mode selection, rate control, etc.  However, for live encoding with low-end devices, such as mobile phones or portable devices, not only is fast encoding speed with reasonable video quality needed but also low power consumption needs to be taken in to consideration.  Power consumption is a critical issue for low-end devices with limited battery power.  In a video encoding system, there are many different components, such as motion estimation, Discrete Cosine Transform (DCT), inverse DCT (IDCT), quantization, motion compensation, etc.  A set of techniques have been developed to provide fast computations and decisions in different components with a tolerable loss of encoding efficiency and video quality for low-end devices with limited power and computation capabilities.

            The set of techniques were developed to make encoding faster for low-end devices using the VC-1 encoder.  VC-1 is the informal name of the SMPTE 421M video codec standard and is a supported standard for HD DVDs, Blu-ray Discs, and Windows® Media Video 9.  The techniques include early termination methods in motion estimation, partial IDCT, optimized picture padding and reduced motion estimation on padding and sliding window subpel refinement.  The new algorithms and implementations together increase the speed of the WinMobile system three times, from ~7 frames per second (fps) to ~25 fps, which makes the encoder more usable in a WinMobile application. 

            There are four algorithms that aid in the increasing of the speed of the fast encoding techniques.  The first one is the quantization parameter (QP) and context/neighbor sensitive thresholds for early termination of motion estimation.  This algorithm can terminate computation expensive motion estimation when the quality of motion compensation is reasonably good.  Therefore, this algorithm can help to save unnecessary computation.  Another algorithm included in these techniques is partial IDCT.  This algorithm takes advantage of the sparse and aggregation property of quantized DCT coefficients and tries to save computations on zero coefficients when videos are encoded at low bit rate.  The third algorithm includes optimized picture padding and reduced motion estimation on picture padding.  This algorithm achieves an optimized compromise between the encoding speed and/or performance, as well as the size of picture padding (e.g. memory usage).  Basically, the algorithm places some constraints on the padding size and motion estimati...