Department of Electrical Engineering, National Tsing Hua University, Hsinchu, Taiwan 300, ROC.
IEEE Trans Image Process. 1998;7(2):238-46. doi: 10.1109/83.661002.
This paper presents a new two-dimensional (2-D) optimum block stochastic gradient (TDOBSG) algorithm for 2-D adaptive finite impulse response (FIR) filtering. The TDOBSG algorithm employs a space-varying convergence factor for all the filter coefficients, where the convergence factor at each block iteration is optimized in a least squares sense that the squared norm of the a posteriori estimation error vector is minimized. It has the same order of computational complexity as another 2-D optimum block adaptive (TDOBA) algorithm. Computer simulations for image restoration show that the TDOBSG algorithm outperforms the TDOBA algorithm and other related algorithms in terms of objective and/or subjective measures.
本文提出了一种新的二维（2-D）最优分组随机梯度（TDOBSG）算法，用于二维自适应有限脉冲响应（FIR）滤波。TDOBSG 算法对所有滤波器系数采用时变收敛因子，其中在每个块迭代中，收敛因子在最小二乘意义上进行优化，以使后验估计误差向量的平方范数最小化。它具有与另一个二维最优分组自适应（TDOBA）算法相同的计算复杂度。图像恢复的计算机仿真表明，在客观和/或主观度量方面，TDOBSG 算法优于 TDOBA 算法和其他相关算法。
