以頻率資訊選擇加強MPEG-4視訊編碼視覺品質

Abstract

在這篇論文裡，以MPEG4視訊編碼規格中的雜訊比可調層次為基礎，我們提出一種更具彈性的概念，可針對不同頻段的資訊個別調整，我們稱之為「頻率資訊選擇」。我們更進一步提出具有這種能力的架構，在位元層編碼前先進行頻段分割，使得各個頻段具有獨立的雜訊比可調層次；然後再針對每個頻段的統計特性設計合適的熵編碼，提升編碼效率。我們並參考人類視覺特性，設計視覺實驗，來決定適當的選擇方式。另外，我們提出的架構，可以經由解碼端進一步的調整，達到「空間解析可調層次」；不同解析度的解碼端，分別接收不同完整性的資料，經由合適的離散餘弦係數調整，以不同尺寸的反轉換還原成不同大小的區塊，組合成畫面後就重建不同解析度的畫面；而移動向量也須作對應的縮小，在不同解析度的畫面執行移動補償。

In this thesis, based on the SNR-scalability scheme in MPEG-4, we propose a flexible approach, which can separately enhance the video quality for different frequency bands. We call this capability “Frequency Information Selection (FIS).” An architecture that provides FIS is also developed. In this architecture, we segment the DCT coefficients into different bands before bit-plane coding so that there is independent SNR-scalability in each frequency band. Then we properly design the entropy codes for each band to improve coding efficiency. Based on human visual perception, we design some experiments to help decide the proper mechanism for the enhancement of visual quality at a given bit-rate. On the other hand, based on the proposed architecture “spatial scalability” is achievable after some proper adjustment at the decoder side. Decoders of different spatial resolutions may receive data up to different frequency bands, scale the DCT coefficients properly, perform the IDCT transform of different sizes, and finally rebuild frames of different sizes. The motion vectors are also scaled down accordingly to perform motion compensation in frames of different sizes.