可適性移動向量搜尋和編碼演算法

Abstract

根據Sandvine公司於2020年進行的調查，全球移動下載流量的65％為視訊。隨著物聯網和5G時代的來臨，高效能的視訊編解碼器對於節省存儲空間和頻帶使用而言，扮演越來越重要的角色。 對於多媒體資料壓縮而言，視訊資料可以達到最高的壓縮率，主要是利用時間上相鄰幀之間的相關性，該技術稱為「幀間預測」。對於預測圖像而言，僅將運動向量和殘差圖像編碼以降低位元速率。然而，在視訊壓縮中，幀間預測也是高計算複雜度的主要原因。 在此論文中，我們提出了透過觀察移動向量的變化來調整搜索區域的演算法，在不影響重建視訊品質和位元速率的前提下，我們試著節省比對次數。由實驗結果發現，我們提出的演算法特別適用於小動作影片或是有特定方向相機運動的影片。 此外，我們也研究了將移動向量編碼的方法。與傳統基於查表方式的編碼方法相比，我們提出了自適性算術編碼，進一步減少位元數。 總結來說，這篇論文回顧了視訊壓縮技術，並且對於移動向量提出搜索和編碼的相關演算法，做實驗驗證想法並討論其提升效能的可能性。

According to the survey from Sandvine in 2020, 65% of the downloaded data are video files. As 5G networks are deployed, efficient video codecs are essential for saving the requirements of storage and internet bandwidth. Among all kinds of multimedia, video can achieve the highest compression ratio due to the high correlation between adjacent frames. The technique of “Inter prediction” makes good use of interframe redundancy. For a predicted frame, only the motion vector and the residual image are coded to bitstream. However, in video compression, inter prediction usually leads to high computational complexity so we believe that there is still room for improvement. In this thesis, we propose an algorithm for adjusting the search window by observing the change of global motion vectors. We minimize the computation without affecting the quality of the reconstructed video and the bitrate. The proposed algorithm is especially suitable for static and fix camera slow movement. Besides, we investigate techniques to encode motion vector data into bitstream. Compared to the classic table-based coding method, we propose a context-based adaptive arithmetic coding (AAC) to further reduce the bitlength. All in all, we review the techniques in video compression and propose efficient methods for searching and coding motion vector.