PCA-ICA neural network model for POLSAR images analysis
Salim Chitroub · 2004
The POLSAR images are modeled by a mixture model that results from the product of two independent models, one characterizes the target response and the other characterizes the speckle phenomenon. For scene interpretation, it is desirable to separate between the target response and the speckle. For this purpose, a PCA-ICA neural network model is proposed. Based on its rigorous statistical formulation, a neuronal approach for the simultaneous diagonalisation of the signal and noise covariance matrices using a PCA transform is proposed. The PC images are uncorrelated and having an improved SNR. However, the speckle is a non-Gaussian multiplicative noise, the higher order statistics contain additional information about it. The ICA method is used to separate the speckle from the PC images and provide new IC images that have an improved contrast. The method has been applied to real POLSAR images. The extracted features are quite effective for the scene interpretation.