Gaussian Noise Removal of Blind Source Separation Based on Image Sequence

This paper proposes a noise reduction method of blind source separation based on image sequence. In the image, the signal and the noises can be considered as the independent components, so the multi-frame images can be considered as the multiple linear combinations of one signal and a lot of noises for the noise's randomness. Due to that, the noises can be removed from the sampled multi-images based on blind source separation (BSS). During the separation calculation, the common Gaussian noise is taken as the object to be eliminated, and the nonlinear principal component analysis (NLPCA) is used as the BSS method and the analysis of the noise reduction is based on changing either the noise degree or the image sampling numbers. This proposed algorithm is compared with the multi-frame average (MFA) that is a famous noise reduction algorithm based on image sequence. The analysis results show that the noise reduction properties by using the proposed method will be affected by the noise degree or the image sampling numbers when eliminating the Gaussian noise; and this method can recover the images which are heavily noise polluted, and the suppression effect on strong noises is obviously better than MFA algorithm.