| Video compression is a dependent optimization problem because the coding decision for the current frame determines the set of available rate-distortion operating points of the motion-compensated frame. To minimize the overall rate-distortion cost of a video sequence, this dependency needs to be exploited. In other words, a coded pixel should be a good reconstruction as well as a good reference for pixels in future frames.; Video compression is also a bit allocation problem. The constrained problem of minimizing distortion for a given bit budget is often converted to an equivalent unconstrained one: minimizing a Lagrangian cost function J = D + λR. This basically says that each coding unit should have the same incremental distortion/rate ratio, often quoted as the equal-slope rule. To achieve the best rate-distortion performance, this rule should be enforced for as many coding units as possible.; To realize the potential of the above ideas, a new video coder is proposed based on wavelet transform and set partitioning in hierarchical trees (SPIHT). First, an optimization algorithm is presented to minimize the rate-distortion cost for each macroblock, effectively enforcing the equal-slope rule at the pixel level. The rate-distortion method is then extended to include future dependency by adding distortion terms from future frames, and an iterative algorithm is presented to solve the dependent coding problem. Finally, motion search is jointly optimized by including motion vectors in the cost function, completing an optimization framework which enforces the equal-slope rule for all bits at the macroblock level and pixels across several consecutive frames.; The result is a video compression algorithm which requires no training, no explicit quantization, no floating-point operation for INTER frames, no entropy coding, and allows precise rate control. Compared to baseline H.263, the new video coding system has faster decoding procedures and achieves an improvement up to 1.12 dB in PSNR or 20.6% savings in bit rate for typical sequences used in the video compression community. |
