基于多线性子空间KL变换的可伸缩视频压缩方法  

A Scalable Video Coding Scheme Using KL Transform in Multiple Linear Subspaces

作　　者：刘健[1] 吴飞[1] 姚蕾[2] 庄越挺[1] 

机构地区：[1]浙江大学计算机科学与技术学院,杭州310027 [2]宾夕法尼亚州立大学计算机科学与工程系University Park PA 16802

出　　处：《电子与信息学报》2010年第5期1188-1192,共5页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(60533090,60603096);长江学者和创新团队发展计划资助课题

摘　　要：该文提出一种基于多线性子空间KL(Karhunen-Loeve)变换的可伸缩视频压缩算法。该算法将纹理不同的图像子块投影到多个线性子空间中去,然后分别进行KL变换编码,使得每个子空间内变换系数更接近高斯分布,从而最大限度地在变换编码阶段提高压缩效率。同时,该算法结合多尺度小波分解,实现了质量可伸缩的视频编码,提高了视频压缩率。通过与DCT、小波变换以及多向DCT算法进行比较,证明该算法可获得更好的率失真性能。This paper proposes a novel transform coding method based on the Karhunen-Loeve （KL） transform in multiple linear spaces. This method uses multiple linear subspaces to approximate image signals, and use a KL transform for each linear subspace, so that the transform coefficients in each subspace is close to Gaussian, which effectively de-correlates the coefficients. An SNR-scalable video coding scheme is developed by combining the method with wavelet decomposition. Experiments prove that this scheme accomplishes better coding efficiency than the schemes employing DCT, DWT, and Directional DCT methods.

关 键 词：可伸缩视频编码 KL变换 多线性子空间 

分 类 号：TN919.81[电子电信—通信与信息系统]