摘要
本文利用二维自回归模型规范方程系数矩阵的近似分块Toeplitz特性,提出一种用于估计二维自回归模型参数的递归最小二乘快速算法.它直接以原始图象灰度数据出发作参数估计且不对其作任何边界条件假设,而现有算法如文[2]中所提出的算法则假定二维相关系数已知且相应的规范方程系数矩阵为T-T结构(元素为Toeplitz阵的分块Toeplitz阵).另外,本文所述二维自回归模型的支持域为一般的不对称半平面(NSHP)而不仅仅是文[2]中的四分之一平面.本文所提出的算法的运算量与文[2]中的算法一样都是0(m(5/2)),此处m指模型参数个数.所以,本文所提出的算法较现有其它算法更实用.
This paper presents a fast recursive in order least-squares (LS) algorithm to estimate parameters of two—dimensional (2-D) autoregressive (AR) models with a general nonsymmetric half plane (NSHP) support rather than the quarter-plane one of [2], based on the near to block Toeplitz structure of the corresponding normal equation coefficient matrix. The algorithm produces the estimates directly from the image gray level data and does not make any assumptions on the boundary conditions that leads to a more accurate estimate, while the existing algorithms such as ones described in [2] assume that the 2-D correlations are known and the normal equation matrix attains a Toeplitz-Toeplitz structure (block Toeplitz with Toeplitz entries). The computational complexity of the present algorithm is 0(m^(5/2)), same as that of [2], where m indicates the number of the estimated parameters. Thus the present algorithm is a more useful one.
出处
《信号处理》
CSCD
北大核心
1991年第3期153-158,145,共7页
Journal of Signal Processing
