經過壓縮過後的影像序列在錯誤率高的網路傳輸時，所遺失的資料經常導致影像被嚴重的破壞，因此需要有效率的錯誤隱蔽處理來減少傳送錯誤的衝擊。目前已經有許多錯誤隱蔽技術提出，不過，當影像序列中的物體的移動過於快速或複雜時，這些錯誤隱蔽技術總是效率不彰。在本論文中，我們發展出二個適合在動態影像壓縮標準MPEG的時間錯誤隱蔽演算法來解決此問題。首先，本研究將每一個損壞區塊被切割成四個相同尺寸的子區塊，再使用邊界匹配演算法，利用損壞區塊的周圍未損壞區塊的運動向量和像素的訊息來預測失去的運動向量以重建損壞的區塊，但發現需要大量的計算，所以再提出另一個解決方法；另一個方法是在空間域使用類神經網路模型中之自我組織映射圖網路(self-organizing map, SOM)進行隱蔽錯誤，SOM使用盲目聚類方式來達到分群目的，因此SOM在本研究中用於估計和重組損壞區塊的運動向量。由實驗結果得知，本研究所發展的錯誤隱蔽演算法在視覺品質和客觀評估上比起傳統的方式有重大的改進，因此本研究對於容易出現錯誤的網路將十分有幫助。 When transmitted over error prone networks, the compressed video can suffer severe degradation. An efficient error concealment (EC) scheme is essential for diminishing the impact of transmission errors in a compressed video. A number of EC techniques have been developed to combat the transmission errors. However, the techniques are always inefficient when the motions of object in a video are fast or complex. In this thesis, we propose two adaptive temporal EC algorithms to conceal the errors for MPEG-coded video. In the first method, each damaged macroblock is further divided into four sub-blocks with equal size. The information of undamaged motion vector and pixels surrounding a damaged macroblock is used to estimate the lost motion vectors of the sub-blocks based on the boundary matching technique. The estimated motion vectors are used to reconstruct the damaged macroblock by exploiting the information in reference frame. However, this approach entails a considerable amount of processing complexity at the decoder. Thus, we propose another method perform high computational efficiency and good visual quality. The second method is an adaptive EC algorithm that conceals the error for macroblock-based coding systems by using neural network (NN) techniques in the spatial domain. In the proposed algorithm, the self-organizing map (SOM) is used to estimate and reconstruct the lost motion vectors of damaged blocks. The SOM has a great capacity for visualizing and interpreting high-dimensional data sets. Simulation results show that the visual quality and the PSNR evaluation of reconstructed frames are significantly improved by using the proposed EC algorithms. From the experimental results, we find that the proposed algorithms are expected to be useful EC algorithms for motion vector compressed video in error-prone networks.
