Automatic Generation of Seamless Mosaics Using Invariant Features

Prajowal Manandhar,Ahmad Jalil,Prashanth Marpu,Khaled Alhashmi

doi:10.3390/rs13163094

Abstract

The acquisition of satellite images over a wide area is often carried out across seasons because of satellite orbits and atmospheric conditions (e.g., cloud cover, dust, etc.). This results in spectral mismatch between adjacent scenes as the sun angle and the atmospheric conditions will be different for different acquisitions. In this work, we developed an approach to generate seamless mosaics using Scale-Invariant Features Transformation (SIFT). In this process, we make use of the overlapping areas between two adjacent scenes and then map spectral values of one imagery scene to another based on the filtered points detected by SIFT features to create a seamless mosaic. We make use of the Random Sample Consensus (RANSAC) method successively to filter out obtained SIFT points across adjacent tiles and to remove spectral outliers across each band of an image. Several high resolution satellite images acquired with WorldView-2 and Dubaisat-2 satellites, and medium resolution Sentinel-2 satellite imagery are used for experimentation. The experimental results show that the proposed approach can generate good seamless mosaics. Furthermore, Sentinel-2’s level 2A (L2A) product surface reflectance data is used to adjust the spectral values for color consistency.

Highlights

IntroductionSatellite imagery is used in several earth observation applications such as land monitoring, classification, object detection, damage assessment, etc
This process is followed by another application of the Random Sample Consensus (RANSAC) method in each band to remove spectral outliers, after which a linear function for each band is developed to map the adjoining tile onto the base tile
We have developed an approach to generate seamless mosaics based on Scale-Invariant Features Transformation (SIFT) features by filtering outliers based on a robust approach

Summary

Introduction

Satellite imagery is used in several earth observation applications such as land monitoring, classification, object detection, damage assessment, etc. Such applications often span across a wider area and at times country-scale or even larger. This area of interest cannot always be captured in one image due to the limitation in the field of view of satellite sensors and the satellite orbits. It is necessary to generate a large-scale composite image with a sequence of overlapped satellite scenes or tiles, which is usually referred to by a technique called image mosaicking [1]. Image mosaicking consists of five aspects which include image registration, extraction of overlapping areas, radiometric normalization, seam-line detection, and image blending [2]

Methods

Results

Conclusion