利用可變動的影像群大小達成基於影像內容決定的可調性編碼分類

Abstract

視訊編碼的可調性是視訊標準H.264/AVC上新納入的一項技術，其包含時間域，空間域，以及視訊品質三種方式，可以滿足許多視訊傳輸上的需求。可調性的特徵是能夠適應多變的傳輸頻寬，以達到在目前許可的條件之下可達到的最佳視訊品質；而三種可調方式的妥善運用可以使得可調性的編碼方式更具有彈性及全面。本篇論文針對一視訊影像的內容作分類，並根據此分類來決定使用何種可調方式進行編碼；適當的分類將會使得可調性的功能完整地發揮，如此一來當在傳輸過程中發生資料丟棄時，仍能因可調性與影像內容的配合而得到良好的影像回復效果。一視訊影像的內容我們將其分為四種類型，第一種是畫面簡單且物體移動慢，第二種是畫面簡單且物體移動快，第三種是畫面複雜且物體移動慢，第四種是畫面複雜且物體移動快。如此的分類是基於人類視覺上，若針對一個影像作縮放，許多細節將會遺失，因此畫面複雜的影像若使用空間域的可調編碼是不適合的；又，當一影像片段包含快速移動的內容時，若使用時間域的可調編碼將不利於遺失影像的回復。實驗結果中顯示，如此的分類使得傳輸且部分遺失的視訊影像能夠擁有較好的回復品質，且不會因為影像內容的變化而造成視訊品質不穩定的現象。

Scalable video coding (SVC) is the extension of H.264/AVC standard. In SVC, it provides three types of scalability, including temporal scalability, spatial scalability, and quality scalability, with which great flexibility could be provided to fit different needs of transmission. Scalability is a great feature for transmitting compressed video data adaptively in variant bandwidth. However, manipulating the three types of scalability in different ways to encode the video sequence can yield a wide range of quality and coding efficiency. This paper exploits the characteristics of different scalabilities on the encoding of video sequence with varying features, in order to achieve satisfactory video quality at the desirable bit-rate. This work focuses on how to dynamically decide GOP size and the coding type adaptively according to the features of video content. With adaptive GOP, the most suitable scalable coding type can be applied for each GOP, and the quality under low bandwidth can be satisfying. We classify the content into four types, including simple frame with slow motion (SS), simple frame with high motion (SH), complex frame with slow motion (CS), and complex frame with high motion (CH). The idea is based upon the human perception that simple frame can present better III visual quality than complex frame after scaling from low resolution to high one, and thus make it a good choice for coding in spatial scalability; on the other hand, it is easier to reconstruct a lost frame with acceptable quality if the lost frame is in a slow motion sequence than in a high motion one, and thus make it a good choice for using temporal scalability. Under good consideration of the content of the video sequence, a better quality can be obtained in low bandwidth compared to standard scalable coding way. In high bandwidth, the algorithm still can adaptively choose the most suitable scalable coding technique for the video sequence. Therefore it works well in variant bandwidth.