A survey of multispectral image denoising methods for satellite imagery applications

Abstract

In comparison with the standard RGB or gray-scale images, the usual multispectral images (MSI) are intended to convey high definition and an authentic representation for real world scenes to significantly enhance the performance measures of several other tasks involving with computer vision, segmentation of image, object extraction, and object tagging operations. While procuring images form satellite, the MSI are often prone to noises. Finding a good mathematical description of the learning-based denoising model is a difficult research question and many different researches accounted in the literature. Many have attempted its use with the application of neural network as a sparse learned dictionary of noisy patches. Furthermore, this approach allows several algorithm to optimize itself for the given task at hand using machine learning algorithm. However, in practices, a MSI image is always prone to corruption by various sources of noises while procuring the images. In this survey, we studied the past techniques attempted for the noise influenced MSI images. The survey presents the outline of past techniques and their respective advantages in comparison with each other.