A Study of Variable Block Size Error Concealment Scheme for H.264/MPEG-4 AVC Coding

[[abstract]]在視訊傳輸上影音資料藉由壓縮技術以減少位元率使資訊能在有效頻寬中傳遞，H.264/AVC視訊壓縮標準相較於先前標準擁有降低位元率的最佳效能，其中一個機制是在動態預估及動態補償中提供了可變的7種區塊模式和5個參考畫面，來改善壓縮率與畫面品質。由於H.264/AVC主要目標是提高壓縮率，對於位元流的抗雜訊能力是比較不足的，且由於在高壓縮編碼效能上每個位元都包含更多的資訊，單一個位元遺失時可能導致解碼端整個區塊的損害或嚴重的破壞視覺品質。為了使解碼端能提升影像品質，H.264/AVC參考軟體採用了錯誤修補技術，但卻未針對不同區塊尺寸大小的錯誤修補上的技術加以探討，依舊會造成影像品質提升有限的問題。 因此，本論文提出兩個有效多區塊大小之錯誤修補方法，來改善錯誤的視訊影像品質。方法一，使用仿射模型(Affine Model)，將擷取到的特徵點與匹配點，以仿射模型近似出曲線，透過LSE準則解出一組參數，利用這組參數來重建遺失的移動向量。方法二，提出最佳權重內差法(Optimal Weighting Interpolation)，透過MSE準則與有效區塊移動向量之垂直與水平方向的內差誤差來預測出最佳的內差權重，並搭配周圍正確區塊之移動向量來重建遺失的移動向量。模擬結果顯示，在影像品質修補上，方法一與方法二修補結果可提高0.9~1.5dB，在物件邊緣上有很明顯的修補效果。[[abstract]]H.264/AVC standard has the best bit-rate reduction comparing to the previous video coding standards. Notably, since a bit in high compression coding contains much more information than it does in low compression coding, any single bit lost in high compression coding may result in the whole block’s loss and serious visual quality degradation for the decoded image at the receiver. If the error is not removed, the decoded image is still impaired. Therefore, this thesis proposes two error concealment algorithms for variable block size scheme to improve corrupted video images. First one, this thesis uses affine model to describe the relationship between corresponding feature points, then the affine parameters is determined by least squares estimation technique. The other one, this thesis proposes an optimal weighting interpolation method to recover the lost motion vectors. The weights are estimated by the technique of mean squared error minimization where the error is raised from the horizontal and vertical interpolations for the lost motion vectors. The simulation shows that, by using the proposed algorithms, the image recovery performance is significantly improved when the video sequence in transmission is corrupted. According to PSNR values, the proposed algorithm is superior to the other methods about 0.9~1.5dB.[[note]]碩