摘要
本文探讨了一种新的多元统计分析方法——独立分量分析在图像边缘特征提取方面的应用.采用基于信息最大算法的无监督神经网络对自然图像进行迭代学习, 获得ICA所需的基函数。提取的基函数在空间频率上具有方向性和局部性,很好地描述了输入自然景物图像的边缘特征。实验结果表明,即使在有噪声的条件下,ICA也可以较好地获得图像的边缘特征信息。
The application of a new multidimensional signal processing method-Independent Component Analysis to image edge feature extraction is discussed in this paper. By using the information maximization algorithm of ICA implemented by an unsupervised learning neural network, it can get the basis functions of the natural images. The basis functions have directionality and locality at spacial frequences and can describe the edge features of the natural images well. The experimental results have shown that the ICA can extract the edge features of images even in the presence of noise.
出处
《红外》
CAS
2006年第5期13-16,共4页
Infrared
关键词
独立成分分析
信息最大化
稀疏性
滑动子窗口
independent component analysis
information maximization
sparseness
sliding subwindow