Modified Neighborhood Similar Pixel Interpolator Approach for Removing Thick Clouds in Landsat Images

Thick cloud contaminations in Landsat images limit their regular usage for land applications. Based on the assumption that the neighboring spectral-similar pixels outside cloudy patches have similar temporal change patterns to the cloudy pixels, this paper presents an improved neighborhood similar pixel interpolator (NSPI) approach to build a cloud-free imagery. NSPI approach was originally developed and tested for filling gaps due to the Landsat ETM+ Scan Line Corrector (SLC)-off problem. Both simulated and real cloudy images were used to evaluate the performance of the proposed method. The results show that NSPI approach can restore the reflectance of cloud-contaminated images with fewer artifact edge effects comparing to a contextual multiple linear prediction (CMLP) method. The reflectance restored by NSPI approach is more accurate especially when the cloud-free auxiliary image and cloudy image are acquired from different seasons and have different spectral characteristics.