Abstract: Reducing the noise in hyperspectral (HS) images to enhance image quality has been a hot research field of remote sensing image processing, but hybrid optoelectronic noise is difficult to precisely estimate. A new method for the HS hybrid noise estimation based on the homogeneous regions (HR) segmentation is proposed. The method makes HR segmentation by combining with spatial and spectral characteristics of the HS image first, and then removes correlation in regions by using multiple linear regression (MLR) to get the hybrid noise. Finally, the scale factor is introduced to estimate the internal parameters of the hybrid noise. The performance of this method is analyzed on simulated HS data and also applied to a well-known airborne visible infrared imaging spectrometer (AVIRIS) data. The experiment demonstrates this method improves the accuracies of segmentation and hybrid noise estimation when compared to other approaches.

Key words: hyperspectral image, homogeneous regions segmentation, hybrid noise, noise estimation