We propose an application that utilises audio and video datadependencies to achieve additional video compression in low-bit rateencoding systems such as: H.263+ video coding and G.723.1 audio codingstandards. The joint correlation of synchronized audio and motionparameters has been proved to exist. A joint performance of principalcomponent analysis (PCA) by Karhunen-Loeve expansions (KL) andtree-structured vector quantization algorithms (TSVQ) based onLinde-Buzo-Gray (LBG) and competitive learning (CL) techniques achieveas much as 60% bit reduction for the motion in the mouth region (1% ofthe overall output bit rate of a P frame) and provide the samemotion-compensated image quality in high picture formats. We showperformance evaluations that determine the optimal audio parameters,such as linear predictive coefficients (LPC) or line spectrum pairs(LSP), and determine the nature of the motion parameter in eachmacroblock of the mouth region when using advanced prediction mode (APM)video coding
展开▼
