建構於MPEG-4之強韌性細粒狀可調層次式視訊編碼技術之參數設定研究

Abstract

在這篇論文裡，以RFGS的架構作為基礎，我們提出一種依據影像特質，針對一段給定的頻寬，快速計算視訊編碼之□與□參數之最佳設定方法。在論文中，將對先前以PSNR為比較基礎的參數最佳選擇法作改善，改成以MSE為比較基礎，並以此為我們提出之快速參數選擇法的目標。首先，我們分析了不同影像在運動補償後的residual特性以及其和最後編碼大小的關係，並發現其規律性；另外我們也歸納出參數的調整在不同的頻寬條件下，與影像品質之間的關聯性。最後依照先前整理出的各種規律性，先建立起在單一頻寬時的參數預測模型，再進一步發展成考量一段給定頻寬的參數最佳化方法。我們提出的方法是利用簡單的運算即可對於不同的影像進行適當的編碼，以畫面(FRAME)為基礎，產生最佳的參數組合。

In this thesis, based on the RFGS coding scheme, we propose an approach to automatically calculate the optimized □ and □ parameters for videos of different properties. Starting from the previous PSNR-based work, we propose new MSE-based approaches to pursue better video quality. We first analyze and model the residuals of motion compensation with different parameter setting under various bit rates. We then investigate the relationship between error drift and available bandwidth, and the relationship between MSE gain and extra bandwidth. With those relationships and models, we proposed first an approach to automatically choose the sets of optimized parameters for a fixed bit rate. Then we proposed a solution for the case with a range of bit rates. Frame-based approaches are studied and these proposed MSE-based approaches require only light computational loads.