Abstract
Accurate approximation of the signal-independent (SI) and signal-dependent (SD) mixed noise from hyperspectral (HS) images is a critical task for many image processing applications where the detection of homogeneous regions plays a key role. Most of the conventional methods empirically divide images into rectangular blocks and then select the homogeneous ones, but it might result in erroneous homogeneity detection, especially for highly textured HS images. To address this challenge, a superpixel segmentation algorithm is proposed in this paper, which can decompose a noisy HS image into patches that adhere to the local structures and hence persist in homogeneous characteristic. A novel spectral similarity measure is defined in the frequency domain to make the superpixel segmentation algorithm more robust to the mixed noise. Combined with an improved scatter-plot-based homogeneous superpixel selection and a multiple linear regression-based noise parameter calculation, our method can accurately estimate SD and SI noise variances from HS images with different noise conditions and various image complexities. We evaluate the proposed method with both synthetic and real Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) HS images. Experimental results demonstrate that the proposed noise estimation method outperforms the state-of-the-art methods.
Highlights
Hyperspectral (HS) imaging is of growing interest as a key technique to Earth remote sensing
We will mainly focus on the random noise, whose elimination is still a challenge; fixed pattern noise can be removed by calibration routines or destriping approaches [10]
Regarding the improvement of the sensitivity in electronic components, recent studies find that the additive noise assumption is no longer appropriate for HS images acquired by a part of new-generation HS spectrometers, where it is more proper to model the noise as a mixture of signal-independent (SI) and signal-dependent (SD)
Summary
Hyperspectral (HS) imaging is of growing interest as a key technique to Earth remote sensing. SSDC estimates spectral and spatial correlation coefficients via the multiple linear regression (MLR) model, and the remaining residuals are considered to be noise components This method performs well on weakly textured HS images but always produces inaccurate noise estimates as the image textures increases, which is mainly due to the lack of an effective homogeneous region detection step. An advanced noise estimation method was proposed in [25], where the homogeneous regions are automatically detected based on the classification of intensity variances of image blocks This method produces accurate mixed noise estimation results, especially for the weakly textured images. The HS image is empirically divided into regular rectangular blocks and the homogeneous blocks are selected with different techniques, which may result in erroneous homogeneity detection, especially in the case of richly textured images To address this challenge, we design a novel superpixel segmentation algorithm (SSA) for the mixed noise estimation. Several experiments have been conducted with both synthetic HS data and real Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) images to analyse the performance of the proposed method, and the comparison with the state-of-the-art methods have been performed
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
