Blind source separation of hyperspectral images in DCT-domain

Abstract

In this paper, we consider the problem of blind image separation by taking advantage of the sparse representation of the study images in the DCT-domain. Blind source separation (BSS) is an important field of research in signal and image processing. The BSS problem has been considered either directly in the original domain of observations or in a transform domain. The idea behind transform domains is that usually an invertible linear transform restructures the signal/image values to give transform coefficients more easily to separate. This paper describes a new method for blind source separation. The latter takes advantage of the sparse representation of structured data in large overcomplete dictionaries to separate independent features. Furthermore, DCT exhibits excellent energy compaction for highly correlated images such as hyperspectral images, which permits to reduce significantly the complexity of the separation. For this purpose, we will exploit the redundancy of neighboring pixels and the correlation of adjacent bands by a new source separation approach based jointly on the Blind Source Separation (BSS) and Discrete Cosine Transform (DCT). In this work, we differentiate from the previous works by using a second order source separation criterion in the frequency domain. The extracted independent components may lead to a meaningful data representation which permits to extract information at a finer level of precision. This approach is of utmost importance in the classification process and should minimize the misclassification risk of hyperspectral images.