基於視覺特性之影像編碼位元率控制

Abstract

編碼後的影像品質最終是由人眼判斷的。然而，現今大部分最佳化位元率控制方法使用的誤差衡量標準，是和人眼感知不全然相關的平均方差及其類似的誤差衡量標準。雖然已經有將人眼視覺特性納入考慮所設計的位元率控制方法被開發出來，但其中絕大多數的方法在設計上未考慮位元率和誤差的最佳化。 在本論文中，我們使用結構相似性為影像品質的衡量標準，提出了一個位元率誤差模型，並基於此模型，開發了用於H.264的編碼器的最佳化位元分配與位元率控制演算法。所提出的方法相較於H.264的參考軟體JM，最多可以達到25%的位元率減少。我們更進一步將所提出的位元率控制演算法與基於視覺特性的模式決策方法整合，相較於JM，整合後的方法可以達到32%的位元率減少。主觀的衡量也顯示了我們所提出的位元率控制方法提供了更好的影像品質。

The quality of encoded video is ultimately judged by human eyes; however, most previous optimum rate control schemes use mean squared error and the like as quality metrics that are poorly correlated with human perception. Although perceptual-based rate control incorporating the characteristics of human visual system has been developed, most existing schemes do not take rate-distortion optimization into consideration. In this thesis, we use the structural similarity index as the quality metric for rate-distortion modeling and develop an optimum bit allocation and rate control scheme for H.264. This scheme achieves up to 25% bit-rate reduction over the JM reference software of H.264. We also show that the proposed scheme can be further integrated with the perceptual-based mode decision scheme under the rate-distortion optimization framework to bring the bit-rate reduction over the JM reference software to 32%. The perceptual quality improvement is examined by subjective inspection.