ICA-based image denoising: A comparative analysis of four classical algorithms

Sparse coding is widely used in signal and image processing. Highly related to sparse coding method, independent component analysis (ICA) can be used to build a statistical model for image processing. However, in practice, when used in image processing, the effect or efficiency of different ICA algorithms are not well studied. To fill this gap, in this paper, the image denoising performance of four classical ICA algorithms, namely, two different implementations of basic Fast-ICA, natural gradient algorithm and optimized Fast-ICA are studied. Firstly, assumptions required by sparse coding method and ICA algorithms are briefly introduced. Secondly, feature extraction and image denoising experiments are conducted to compare the performance of different ICA algorithms. The experiment results show that all ICA algorithms mentioned above can be used to explore natural image feature and image denoising, but the results are not always similar. The optimized Fast-ICA algorithm outperforms the other algorithms.