Bilinear models for nonlinear unmixing of hyperspectral images

This paper compares several nonlinear models recently introduced for hyperspectral image unmixing. All these models consist of bilinear models that have shown interesting properties for hyperspectral images subjected to multipath effects. The first part of this paper presents different algorithms allowing the parameters of these models to be estimated. The relevance and flexibility of these models for spectral unmixing are then investigated by comparing the reconstruction errors and spectral angle mappers computed from synthetic and real dataset. This kind of study is important to determine which mixture model should be used in practical applications for hyper-spectral image unmixing.