基於標量量化多描述編碼的錯誤彌補法

Abstract

近年來數位視訊壓縮已成為我們日常生活中不可或缺的一部份。H.264/AVC是一被普遍認可的標準，該標準既可將視訊來源壓縮成較小容量卻又不劇烈犧牲其品質。儘管如此，由於不穩定的傳輸環境，在傳輸視訊串流過程中無可避免地會發生影響視訊品質甚劇的封包遺失與錯誤。因此，在封包遺失與錯誤發生的狀況下勢必得採取一些彌補措施以保證一定的視訊品質。 當無法執行再次傳輸時，透過標量量化多描述編碼可讓視訊串流在傳輸過程中增加錯誤恢復力，保證其視訊服務可提供穩定且可被接受的品質。根據現有研究，大多數的作法對於重建係數的選擇並未考量H.264/AVC量化後係數的機率分布，也沒有考慮不同視訊的差異。本篇論文提出一根據量化後係數的機率分布狀況設計重建係數的選擇方法。實驗結果顯示，當封包無法正確收到及解碼時，此方法可重建出比現有研究更優異的品質。

In recent years, digital video technology has become an important part in our daily lives. H.264/AVC is an approved video compression standard which compresses digital video source to reduce its bitrates but still keep good quality. Nonetheless, due to instable transmission environment, transmitting an encoded video sequence usually confront packet errors or losses that damage video quality severely. Thus, it requires the adoption to guarantee reconstruction quality at the decoder while the information does not delivered successfully. Multiple description scalar quantizer (MDSQ) adds error resiliency to video streaming applications where retransmission is not available. According to the literature, the reconstruction level sets are taken without considering the probability distribution of H.264/AVC residuals. This paper proposes methods to concealment the MDSQ video with proper reconstruction sets. As a result, better PSNR performance can be achieved without introducing additional bitrate.