Improved Visualization Using a Fusion Technique Based on KNN Matting of Remotely Sensed Images

Abstract

Hyperspectral image fusion is an important task in remote sensing as it could strengthen hyperspectral bands for visualization and analysis of satellite images. Many fusion techniques have been developed in the recent years to obtain an accurate and complete description of visualization. Since the existing works could hardly produce the desired results, the proposed method attempts to design a novel fusion framework based on K Nearest Neighbours (KNN) matting model. Image matting aims at finding the probability that each pixel in a band belongs to a specified class. This model capitalizes on the natural principle of matching non-local neighbourhoods by using KNN and contributes a fast and simple algorithm that produces competitive results, provided an input of sparse mark-ups in the form of scribbles. KNN matting has a closed form solution that leverages the existing approaches by producing efficient results. After determining the value of alpha, the solution can be generalized to solve the multilayer extraction problem with reduced computational complexity. Experimental evaluation on benchmark data sets indicates that the proposed model is better than the state-of-the-art methods.