北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2010, Vol. 33 ›› Issue (5): 108-111.doi: 10.13190/jbupt.201005.108.zhangtq

• 研究报告 • 上一篇    下一篇

梯度自适应在线ICA的改进

张天骐1,李雪松2,夏淑芳1,侯瑞玲1   

  1. 1重庆邮电大学 信号与信息处理重庆市重点实验室; 2重庆信科设计有限公司
  • 收稿日期:2009-12-31 修回日期:2010-03-22 出版日期:2010-08-28 发布日期:2010-06-15
  • 通讯作者: 张天骐 E-mail:zhangtianqi@tsinghua.org.cn
  • 基金资助:

    (60602057,10776040);国家级.国家自然科学基金项目

Improved Online ICA Method in Natural Image

  • Received:2009-12-31 Revised:2010-03-22 Online:2010-08-28 Published:2010-06-15

摘要:

为了更好地解决传统梯度下降算法中收敛点难以确定的难题,根据数字图像信号有界的特点,提出一种改进的梯度自适应在线独立分量分析(ICA)算法. 该算法将传统梯度在线算法中的收敛点强加于学习过程,使其由传统的梯度下降过程变为上升过程,保证接收端在最后一组信号到达时,分离矩阵可保持在最优分离点上. 理论分析和仿真结果表明,本算法具有较好的稳态性能和数值稳定性,是一种有效的ICA算法.

关键词: 独立分量分析, 梯度算法, 在线学习, 收敛点

Abstract:

In the conventional gradient algorithm, the convergence points are difficult to be found, an intensive analysis on the gradient adaptive online independent component (ICA) methods is presented. It indicates that the digital image is a magnitude bounded signal, the convergence point in the conventional gradient algorithm will be imposed on the learning process, making the gradient descent processing convert rise processing. It ensures that when the ending codes get to the receiver, the separation matrix can be storage at the optimal point of separation. Simulation shows that this new method is with stable performance and numerical stability, and is an efficient independent component analysis algorithm.

Key words: independent component analysis, gradient algorithm, online learning, convergen ce point