用於畫面之間的小波轉換編碼以人類視覺系統為基礎的位元控制法

Abstract

因為在大多數的應用中，不同的接收者會有不同的承受量，故可調整性(scalability)在今天的多媒體傳輸中是一個重要的特性。用於畫面之間的小波轉換編碼(Interframe Wavelet Video Coding)是一個新的視訊編碼方式且能提供良好的可調整性。因此這個編碼方式在近年來受到不少矚目，而且已經有很多的研究和改良來增進它的效能。 在很多環境下，人眼都是視訊品質的最後判斷所在。然而，在設計視訊編碼時要包含人類視覺卻很困難。我們必須要能把客觀的“數學上的不同”轉換成主觀的“視覺上的不同”，也就是說，我們必須要把普通的“量化錯誤”轉換成“人類視覺上的加重錯誤”。 在位元控制法(rate control algorithm)中，每個在用於畫面之間的小波轉換編碼的截斷點(truncation point)都有自己相關聯的失真(distortion)和位元長度(bits length)。而每個截斷點的斜率(slope)就是把失真的差異(distortion difference)除以位元差異(bit difference)所得到的商。在最佳化理論中(optimization theory)，擁有較高斜率的截斷點有較高的優先權被傳送。在本論文中，我們提出一個方法，就是說我們把每個截斷點的斜率乘上一個由人類視覺系統算出來的比重。故這個經過視覺加重的斜率會成為位元控制法中判斷的標準。我們的模擬會指出最後的重建影像有較低的最高訊號雜訊比(PSNR)和較佳的視覺品質。

Scalability is an important feature in today’s multimedia transmission because in many applications receivers have very different capabilities. Interframe wavelet video coding is a new video coding algorithm that can achieve fine-scale scalability. Therefore, it has received a lot of attention recently and many research and development projects have been conducted to improve its performance. For most entertainment purposes, human eyes are the final judge of the video quality. However, it is rather sophisticated to include the human perception in the video codec design. We need to transform the objective “mathematical difference” into the subjective “visual difference”, i.e., we need to convert the ordinary “quantization error” to the “human-visual weighted error”. In the rate control algorithm, each truncation point in the interframe wavelet video coding has its associated distortion and bits length. The slope of each truncation point is the quotient of the distortion difference divided by the bit difference. Based on the optimization theory, the truncation point with a larger slope should have a higher priority to transmit. In this study, we propose a method that we weight the truncation point slope by a weighting factor, which is derived based on the human visual system. Thus, the visually-weighted slopes become the criterion in rate control. Our simulations indicate that the reconstructed frames may have lower PSNR but higher visual quality.