具可適性 GOP之 H.264/AVC視訊壓縮平行化處理

Abstract

與MPEG-2、H.263或MPEG-4相比，H.264/AVC是一個效能更佳的視訊壓縮標準，H.264/AVC可以讓視訊檔案變更小，同時保有相似的品質。一般而言，圖片群(Group of pictures)的內容會影響壓縮效能，所以圖片群的尺寸選取應該要根據不同內容變化達到可適性調整。本篇論文將基於場景變換偵測來動態調整圖片群尺寸。此外我們提出了兩種在幀級(Frame-level)上的平行化排程，由於導入可適性的圖片群尺寸決策，將使每個圖片群的尺寸不再一致，這也將提高平行化排程的挑戰性。根據實驗結果，與固定尺寸的圖片群相比，我們的方法可以得到最佳的比率-失真率(Rate-distortion)。至於平行化處理的效能，我們的方法可以有效地平衡各線程(Thread)間的編碼時間，藉此達到較佳的加速效果。

H.264/AVC is a standard for video compression which obtains better performances relative to former standards such as MPEG-2, H.263 or MPEG-4. H.264/AVC can compress video into smaller size and maintain the similar video quality. In general, coding performance is influenced by the content variations in group of pictures (GOP). Therefore, the GOP size determination is a critical issue for video coding, it should be adaptively adjusted according to different video content variations. This paper exploits this adaptive GOP determination based on scene change detection. Besides we proposes two kinds of frame-level parallel scheduling. Because the application of adaptive GOP determination, i.e. non-constant GOP size, it makes the proposed parallel scheduling more challenging. According to our experimental results, the adaptive GOP determination obtains better rate-distortion than the fixed GOP size. As for the performance of parallel scheduling, the proposed methods can balance the encoding time among threads, and can achieve the better speedup.