Multispectral image denoising in wavelet domain with unsupervised tensor subspace-based method

Multiway Wiener filtering has been inserted in a wavelet framework to enhance spatial details while denoising multidimensional images. An elevated number of rank values is required. A solution is to retrieve the best rank values while minimizing a mean square criterion. In this paper, we justify the adaptation for this purpose of a stochastic optimization method, and we evaluate comparatively a genetic algorithm and particle swarm optimization. Results obtained on multispectral images in terms of signal to noise ratio and perceptual image quality permit to emphasize the performance of the obtained unsupervised method for realistic noise magnitudes.