Matrix decomposition based salient object detection in hyperspectral imagery
Yifan Gao, Hangqi Yan, Lei Zhang, Runping Xi, Yanning Zhang, Wei Wei · 2017
Salient detection in hyperspectral images (HSIs) can be benefited by the abundant spectral information. Most related methods adopt integrating the spectral characteristics into the traditional Itti's model to consider the local region contrast. However, these methods often segmente the object into several pieces and are sensitive to uneven illumination. To address these problems, we propose a novel matrix decomposition based salient object detection method for HSIs. With being modelled with spectral gradient feature, the HSI is decomposed into a low-rank background matrix with a sparse one which can indicate the salient object with more intact appearance. In addition, the spectral gradient feature guarantees the proposed method to perform robustly with uneven illumination. Experimental results demonstrate the effectiveness of the proposed method.