Rate-distortion Optimized Mode Selection Based on Multi-channel RealizationsOverviewMotion-compensated hybrid coderProposed Improvements over H.263Multiple Channel RealizationsChannel RealizationsError ConcealmentDistortion MeasureMode SelectionChannel DecoderExperimental SetupPerformance Measurement (I)Performance Measurement (II)Number of RealizationsError Propagation (I)Error Propagation (II)ConclusionsRate-distortion Optimized Mode Selection Based on Multi-channel RealizationsMarkus GärtnerDavide BertozziClassroom Presentation13th March 2001Markus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 20012OverviewHybrid Video CodingProposed ArchitectureMulti-channel realizationsPerformance Measurements:Concealment TechniquesNumber of Channel RealizationError PropagationConclusionsMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 20013Motion-compensated hybrid coderIntraframe DCT coderMotion compensated predictorIntraframeDecoderMode ControlXORDecoderEncoderMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 20014Proposed Improvements over H.263H.263 Hybrid Video Coder:Error propagation in the decoder loop neglectedError-free transmission assumedThreshold based mode selectionGoals of our approach:Simulation of several channel conditionsPrediction of the error incurred at the receiverRate-Distortion optimized mode selectionMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 20015Multiple Channel RealizationsCoderFrame BufferDecoderInterIntraChannel nChannel nDecoderDecoderConceal-ment & Mode DecisionOriginal Encodernth Channel RealizationinputoutputMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 20016Channel RealizationsRandomly generated error patterns for each channel Capture different sensitivity of macro-blocks to errorsChannel 1Channel 2Channel nXXXXXGroup of blocks (GOB)Estimate of the real channel conditions (on the average)Markus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 20017Bit error causes loss of macro blockSynchronization markers before each GOBMacro block concealmentGOB concealmentConcealment of rest of GOBError ConcealmentXXXXXXErroneous macro-blocks are replaced by respective macro-block of previously reconstructed frameMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 20018Distortion MeasureChannel 1Channel NMB tedreconstruc - MBcurrent 1dMB tedreconstruc - MBcurrent NdFor each Macro-block:NiidND11Markus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 20019Mode SelectionDecision takes place for each macro-block selection tableComputational complexityInput FrameINTERINTERINTERDRJINTRAINTRAINTRADRJ><Mode SelectionMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 200110Channel DecoderSelection tableCoderModeinputIntra: quantized frameInter: difference signal, motion vectorsFor each channel:Inter: previous frame buffer content + difference signalIntra: quantized frameReconstructed FrameMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 200111Experimental SetupEncoderDecoderChannelQuantizerFrame BufferFrame BufferDequantizerDistortionRateMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 200112Performance Measurement (I)Markus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 200113Performance Measurement (II)Markus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 200114Number of RealizationsMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 200115Error Propagation (I)First I-Frame received in errorMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 200116Error Propagation (II)First I-Frame received correctlyMarkus Gärtner, Davide Bertozzi: Robust Video codingStanford University, 13th March 200117ConclusionsSuitability for error-prone environmentsBetter performance than H.263Reduction of error propagationLimitationsAdvanced modes of H.263 not consideredComputational complexityApplication for downloadable multimediaFuture work:Sophisticated channel modelsImplementation of advanced
View Full Document
Unlocking...