Approximated Iterative QLP for Change Detection in Hyperspectral Images

We propose a matrix factorization algorithm based on the iterative Stewart’s QLP decomposition. In particular, provided a given threshold, only an automatically selected subspace is used to approximate the original dense matrix. The algorithm is validated on the change detection task for Hyperspectral Images (HSI). The extraction of information from HSI is an important field of research relevant to many applications. In the aerospace sector, for example, it is useful to monitor changes of the Earth surface, or to find salient information from urban geo-spatial data. Therefore, low rank approximation techniques play a fundamental role.