Automatic cloud and cloud shadow removal method for landsat TM images

Abstract

Optical Remote Sensing Images are often interfered by clouds and their shadows. In this research, a scheme is proposed to automatically detect and remove clouds and their shadows by integrating complementary information from multitemporal images to generate the cloud-free composite images. Firstly, image classification is used to separate cloud regions and shadows regions of input images, and shadow regions can be revised by watershed detection method based on NDVI. Secondly, vegetation phenology characteristics of base image are applied to that of reference images where the complementary information is extracted from. Thirdly, fusion method based on multi-resolution pyramid is adopted for smoothing the mosaic artifacts. Finally, evaluation of pixels reliability is proposed to distinguish various sources of the composite image.