利用連續式量化架構實現Wyner-Ziv編碼

Abstract

Wyner-Ziv 編碼為一失真壓縮技術，其利用解碼器所能得到的附加資訊(side information)來幫助解碼。我們提出一個能實現的編碼設計方法。同時我們也根據連續式量化架構提出一個量化演算法以用來實現二次高斯(QuadraticGaussian) Wyner-Ziv 編碼。我們使用的編碼架構會使用兩組獨立的成分編碼器(component code)。而利用隨機編碼(random coding)分析，這個編碼架構已經被證明可以達到Wyner-Ziv 邊界(Wyner-Ziv bound)。 我們會利用能夠趨近通道容量(capacity)的通道編碼(channel code)以及能夠趨近資料率失真邊界的訊號源編碼(source code)來實現我們提出的編碼設計方式。和現存的箱式(binning)編碼設計比起來，我們提出的編碼設計方法以及量化演算法有著合理的複雜度以及良好的表現。

Wyner-Ziv coding is a lossy compression technique which uses the decoder side information to help reconstruction. We propose a practical code design methodology and a quantization algorithm inspired by successive quantization to solve the Wyner-Ziv problem in quadratic Gaussian case. Our coding structure uses two independent component codes and this coding structure has been proven to achieve the Wyner-Ziv bound by the random coding analysis. We also implement it with existing rate-distortion-bound-achieving quantizer and capacity-achieving channel code. Compared to the existing binningbased design approaches, our design methodology and quantization algorithm have reasonable complexity with good performance.