MPEG-2 Error Concealment Based on Block-Matching Principles Sofia Tsekeridou and Ioannis Pitas IEEE Transactions on Circuits and Systems for Video Technology, June 2000
Types of Errors Bitstream errors: –direct bit errors or packet loss in bitstream Propagation errors: –caused by predictive coding. Such as temporal prediction in applied in P- and B-frames
Best Match Region similarity evaluation –MAD smaller than a predefined threshold Statistical behavior evaluation –Difference of Top and Bottom mean must in between Q*variance (Q set to 0.5)
Error Concealment in this paper Spatial method –Split-match EC method (for intra-frame concealment) Temporal method –forward-backward block-matching EC method (F-B BM EC)
First Intra-Frame EC Bilinear Interpolation –From Existing Top and Bottom MB T, MB B Spatial Anisotropic Diffusion EC –Diffusion of the surrounding image content to the lost areas with edge preservation Spatial Split Match EC Method
Spatial Split-Match EC Method copy
Spatial Split Match EC Method Step1: Maximum Block Size (16x8)-Copy Step2: Vertical Directions Only (First Split 8x8) –Copy Step3: Smaller Set of Directions (Second Split 4x4) – Diffusion or interpolation Step4: Larger Set of Directions (Enlarge search regions) - Diffusion or interpolation
Intra-/Inter-Frame EC Zero motion EC (ZM EC) Motion-compensation EC (MC EC) –Average neighboring motion vectors (intra coded frame has no MV->ZM) Adaptive EC –When spatial activity is high, apply temporal MC EC ;when temporal activity is high, apply spatial interpolation.
Experiments
Flower Garden PER=0.01
Flower Garden PER=0.1
Mobil & Calendar PER=0.01
Mobil & Calendar PER=0.1
Tunnel PER=0.01
Tunnel PER=0.1
SD EC SVI EC
Adaptive ECF-B BM EC
MC EC
Adaptive ECF-B BM EC
MC EC
Conclusion Appling the spatial split-match EC method to the first I-frame of a sequence or a scene and uses the temporal F-B BM EC method to the other frames. Succeeding in reconstructing edges or lines of various orientations, textures and other spatial details in a smooth and continuous way.