摘要
基于格型矢量量化克服了传统矢量量化计算和存储复杂度高的缺点,提出了一种新型四叉树结构的小波格型矢量量化算法。对小波变换系数进行视觉加权,以感知的均方误差准则取代传统均方误差。改进零树编码方法,并按照小波系数的四叉树结构和其方向性来组织矢量,对重要系数作格型矢量量化。从而较好地综合利用了小波系数在空间和频域的能量集中特性。仿真实验证明,该方法具有速度快,存储量小的优点,与文献中其他算法相比,具有较高的编码效率。
Lattice vector quantization（LVQ）outperforms LBG based vector quantizers by offering a substantial reduction in computational and storage complexity. A novel algorithm based on quadtree structure is proposed in this paper as a combination of wavelet transform and LVQ. First, perceptual weighting on the wavelet transform coefficients is introduced with the result of substituting normal distortion measure Mean Square Error（MSE）with perceptually based one. To exploiting both frequency and spatial compaction of energy of wavelet, a framework with improved zerotree coding followed by LVQ is developed. Codeword is organized based on directional quadtree structure to take advantages of the inherent properties of the wavelet transform coefficients. Stimulation shows, compared with other algorithms in literature, this algorithm has distinct advantages of high coding efficiency with fast speed and small storage requirement.
出处
《电子科技大学学报》
EI
CAS
CSCD
北大核心
2005年第4期460-463,共4页
Journal of University of Electronic Science and Technology of China
关键词
格型矢量量化
四叉树
视觉加权
零树编码
lattice vector quantization
qundtree
perceptual weighting
zerotree coding
