基于自适应学习率独立分量分析的图像盲分离

Blind separation of images using adaptive learning rate independent component analysis

以独立分量分析(ICA)技术作为主要研究对象,对基于独立分量分析的定点算法进行了详细的分析和推理。传统定点算法具有结构简单、运算速度快的特点,但是在图像盲分离中数据有时不能完全满足独立性假设,因此在有些情况下,该算法是否收敛仍具有不确定性。由此,提出了一种能够自适应调整学习率的改进定点图像盲分离方法。将该方法用于混合图像的分离中,较传统方法而言,有收敛速度更快、鲁棒性更强、对数据相关性要求相对较低的优点。计算估计图像的峰值信噪比可知,分离效果是十分有效的。可见,该算法是一种新的、快速有效的图像分离方法。

It is the technique of Independent Component Analysis（ICA） that be mainly researched in the paper.The Fast Independent Component Analysis has the advantage of sample structure,fast convergence,easy to use and so on,but it requests the data independent of each other rigorously.In fact,there are many relationships in the real observation data sets.So,when these things happen,traditional Fast ICA convergences very slowly even fail.We solve the problem by proposed an adaptive learning rate to improve the convergence performance.The experiment is applied to image separation and compared to the traditional methods in the convergence speed.It is easy to see this method is very efficient after calculating the Peak Signal Noise Ratio （PSNR） of the obtained images.