Fast Noise Estimation in Hyperspectral Images Using Spectral Absorption Parameters

Estimation of noise present in Hyperspectral images is a way to enhance the quality of the extracted information and to reduce the uncertainties in the results. The simplest method widely used in noise estimation is Shift Difference. This method has two weaknesses; first, it is based upon the assumption that the adjacent pixels have the same signal information which is not necessarily valid in hyperspectral data sets; second, in order to calculate the correct values of noise it needs homogeneous regions that is usually being determined by supervision. In this study, a new method in noise estimation (NETAL) is introduced. In this method the satellite images are divided into homogeneous regions using spectral absorption parameters such as location of absorption lines, width and depth of these absorption features for every individual pixels. Then in each region the noise was calculated using regression between adjacent bands and finally the total noise was estimated through accumulation of the calculated noises in each region. The NETAL algorithm was evaluated by using simulated and real hyperspectral data sets. The results show that the noise estimation by NETAL method is faster than Multiple Regression method while the accuracy will remain the same as and even better than the Multiple Regression method.