FIRM: Framework for Image Registration Using Multistage Feature Detection and Mode-Guided Motion Smoothness Keypoint Optimization

Remote-sensing image registration is a pivotal preprocessing step for earth observation data analytics. In this regard, georeferencing corrects the systematic geometric degradation in the image. However, it is difficult to achieve subpixel geometric accuracy across multitemporal scenes. This article focuses on hindmost part of geometric correction that uses reference layer and feature detection in hierarchical stages to improve the geometric fidelity of images at subpixel level. The methodology developed is based on patch affine-oriented fast and brief with mode-guided tiled scale invariant feature transform (MT-SIFT) techniques in a coordinate manner at a multistage processing architecture, which we refer to as FIRM. Motion smoothness constraint (MSC) keypoint correspondence optimization is used in FIRM to remove the outliers at gross stage and estimate segmented affine transformation parameters at finer stage. The automatic coregistration pipeline is evaluated in Indian Resourcesat multispectral camera images covering diverse landscapes. The capability of the designed framework is demonstrated to handle relatively large geometrical error. With more than a decade difference in acquisitions, multitemporal images are superimposed over each other and compared with state-of-the-art feature-based methods. The potential of the proposed approach FIRM is assessed on multisatellite imagery acquired from Resourcesat-2 and Landsat −8. It is observed that the root mean square error (RMSE) between coregistered images is 0.12 pixel at a spatial resolution of 5 m.

I. Introduction

Mapping planet earth using remote-sensing data is an enduring research field for scientists around the globe. In this area, Indian Space Research Organisation (ISRO) Resourcesat-2/2A satellites provide vital multispectral information in visible and SWIR wavelength range of electromagnetic spectrum for resource monitoring. LISS-4, LISS-3 and AWiFS Cameras [1], [2] on board Resourcesat spacecraft provides medium resolution multispectral remote-sensing images through its unique three tier imaging concept [shown in Fig. 1(a)]. Table I contains the Resourcesat-2/2A sensors’ specifications. On ground, one of the major data processing steps is to do geometric calibration to determine what location a remote-sensing pixel is actually viewing [3]. The radiometric measurement of a pixel is futile if we do not know where the pixel actually looks. Geometric calibration and correction is a complex job due to resolution, provision of providing tilt to cover region of interest and different acquisition times [4]. The image georeferencing process corrects the system level geometric distortion and establishes the relationship of pixel look angles between image and ground. The geometric modeling using internal sensor parameters (focal length, detector angles, and payload alignment angles) and ancillary data (ephemeris and attitude) improves geometric location accuracy to a certain extent. However, it has been observed that subpixel level geometric registration across multitemporal images is not met, which is a mandatory step to generate analysis ready data (ARD) product [5]. Such georeference data products are burden for the application users to carry out any kind of scientific studies such as change detection, object classification, and time series analysis.TABLE I Resourcesat Sensor Specifications

Sensor	Wavelength (in )	Spatial Resolution (in m)	Swath (in km)
LISS-4	0.52–0.59	5.8	70
	0.62–0.68
	0.77–0.86
LISS-3	0.52–0.59	23.5	141
	0.62–0.68
	0.77–0.86
	1.55–1.70
AWiFS	0.52–0.59	56.0 at nadir	740
	0.62–0.68
	0.77–0.86
	1.55–1.70

Fig. 1.

(a) Resourcesat three tier imaging mechanism. (b) LISS-4 reference layer over India and its surrounding region.