H.264視訊編碼標準提供了高效率的視訊壓縮,目前已是數位電視、行動視訊、與影音串流等各項多媒體通訊服務的關鍵技術。在網路傳輸過程中的封包遺失,會使得影像品質大受影響,也因此有錯誤隱藏技術的發展。傳統的錯誤隱藏技術通常使用固定的區塊大小及樣式,再藉由估測遺失區塊的動作向量內插出遺失的像素;由於影像區塊未必等同於視訊物件,所以其錯誤隱藏的效果也會受到影響。本論文以視訊物件做為錯誤隱藏的基本單元,先在參考畫面進行物件分割,再利用動作向量的一致性加強分割結果;修補時以物件為單元對錯誤畫面進行比對,找出每個物件專屬的動作向量;最後使用EBMA方法檢測使用動作向量之正確性,然後進行貼補。本論文的方法在顏色比較單純而且物件面積較大的情況下可以得到較好的結果,與過去的方法相比平均PSNR可以提升0.02 dB至0.22 dB之間。
H.264 video coding standard, owing to its excellent compression efficiency, has become the key technology for various multimedia communication services such as digital TV, mobile video, and video streaming. Lost data will cause severe visual degradation and error concealment is thus developed for combatting the packet losses during transmission. Traditional error concealment techniques usually use blocks of fixed size and type. The lost data are interpolated by estimating the motion vector of missing blocks. The performance of error concealment is limited because an image block is not equivalent to a visual object. In this thesis, the visual object is used as the basic unit of error concealment. First, object segmentation is performed on reference frames and the uniformity of motion vectors is used to reinforce the result. Second, the motion vector of each object is identified by comparing pixels in reference and current frames. Finally, concealment is done after the motion vector is verified by the extended boundary matching algorithm (EBMA). The proposed method performs well for videos with simple color distribution and large objects. The average PSNR is increased by 0.02 dB to 0.22 dB as compared to the previous method.