雖然最新的H.264視訊編碼提供了很多錯誤防範機制，但是錯誤蔓延所造成的傷害依然很嚴重。本論文提出兩個錯誤防範機制，可以在有錯誤的通道上改善H.264的強健性。這些機制都是和H.264相容的，也就是說位元流可以被標準的解碼端正確的解出。第一個機制是最相似臨近區塊移動補償錯誤隱藏(NNEC)，利用資料隱藏技術藏入移動向量最相似臨近區塊的索引到下一個slice裡。一旦解碼端沒有正確收到MB的移動向量資料，將會利用從位元流解回來最相似臨近區塊的索引來還原，錯誤隱藏效果會比沒有任何資料參考的情況下所做的錯誤隱藏好很多。 第二個所提出的機制是失真量估測模式決策(DEMD)。H.264模式決策是針對碼率失真最佳化，在運算中並沒有考慮到通道錯誤。因此H.264所選到的最佳模式，對有錯誤的環境下並不是最佳的。在所提出來的DEMD裡，會給一個封包遺失率，讓編碼端用遞迴的方式計算解碼端像素的失真量。在解碼端，重建畫面的像素失真量會包括量化失真、錯誤蔓延的錯誤和錯誤隱藏所帶來的錯誤。這些計算都會整合到碼率失真模型裡，去從每一個MB intra和inter編碼模式裡，選出最佳編碼模式。 從模擬結果來看，所提演算法的碼率失真表現在所有的實驗數據下，都較以往的機制好很多，尤其是將NNEC和DEMD整合，更能提供優異的效能。 Although H.264 video coding standard provides several error resilience tools, the damage caused by the error propagation may still be tremendous. This thesis proposes two additional error resilience techniques that can improve the robustness of H.264 in noisy channels. Both of them are H.264 compatible, i.e., the bit streams can be decoded by a standard decoder with little or no degradation. The first technique is Nearest Neighbor motion compensated Error Concealment (NNEC) that utilizes a data hiding technique to embed the index of the neighboring block with the most similar motion vector into the bit stream, presumably the next slice. Once the decoder does not receive the correct MB data including its motion vector, it can use the nearest neighboring MB with the index extracted from the bit steam to perform much more accurate error concealment than the conventional blind motion concealment. The second proposed technique is Distortion Estimated Mode Decision (DEMD). The mode decision in H.264, which is critical for the rate-distortion performance, does not consider the channel error in the computation. Therefore, an optimal mode selected by H.264 may not be optimal for a noisy environment. In the proposed DEMD, the encoder keeps estimating the decoder pixel distortion recursively for a given packet error rate. The overall pixel distortion of frame reconstruction at the decoding side includes the quantization error, the propagation error, and the concealment error. The estimates are integrated into a rate-distortion model for optimal switching between intra and inter coding for each MB. By observing simulation results, the rate-distortion performances of the proposed algorithms are better than that of the conventional algorithms in all experimental cases. Specifically, the combination of NNEC/DEMD provides excellent performance.
