On Shrinkage-based Edge Detection in a Pca Subspace for Penumbral Imaging

We propose a robust edge extraction algorithm based on Principal component analysis (PCA) for Poisson noise reduction. In the proposed edge detection, the image is firstly transformed to PCA subspace with sparse PCA basis functions and then the noisy components are removed by a soft threshold (Shrinkage). The proposed edge extraction method, which is used as a preprocessing step of the reconstruction, has been successfully applied to penumbral imaging. Both simulation results and real experimental results show that the reconstructed images are dramatically improved in comparison to that with the conventional noise-removing filter.