基於H.264/AVC的混合式多重描述編碼

Abstract

在即時視訊傳輸系統，如視訊會議、點對點視訊串流或是網路電視，時常會有網路連線瞬間中斷導致的視訊品質低落，甚至無法解碼的情形。當錯誤發生時，如果等待重新傳送先前發生錯誤的封包則會使得接收端的播放延誤。多重描述編碼是一個相當適合解決此種問題的系統架構。其特性是不使用重傳機制而將原始資訊平均分散在多個描述子，任一個描述子都可單獨解碼還原，接收的描述子越多則還原的品質越佳。 此篇論文中，我們提出一個基於H.264/AVC，在空間域與頻率域中切割資訊的多重描述編碼架構模型，包含編碼端與解碼端。在編碼端中，對任一個原始視訊，在編碼過程中，會經由空間域的第一次切割產生兩個描述子，接著於頻率域進行第二次切割產生四個描述子。而在解碼端會在解碼過程中將所接收到的描述子合併。當有描述子發生錯誤或遺失時，在解碼時會利用空間域與頻率域上的相關性來達成錯誤隱藏，以增進視訊的品質。由實驗結果可得到，相較於先前之基於空間域方法，在發生描述子錯誤時，可以有較好的表現

In real-time video streaming systems, such as, video conferencing, peer-to-peer video streaming or IPTV, network errors frequently occur and result in degradation of video quality or failure to decode the bit-stream at the receiver. Retransmission of error packets imposes unacceptable delay of the video sequence. Multiple description coding (MDC) is an ideal approach, which generates multiple descriptors with equal importance, to overcome such problem. In this thesis, a H.264/AVC based multiple description coding model, which splits information in spatial and frequency domain, is proposed. In the encoder, two descriptors will be generated after the first splitting in the spatial domain, and then the second splitting is done in the frequency domain for each descriptor, resulting in four descriptors in total. In the case of descriptor loss, the decoder will utilize the correlations in spatial domain and frequency domain for error concealment. The experimental results show that the proposed hybrid model has better RD performance than the existing spatial domain based MDC model.