Synthetic Aperture Radar Image Registration Using Speeded Up Robust Features

Abstract

Geometrically aligning images to a common geometrical axis is called image registration. Registration is required for images of a common scene obtained from different configurations to be used in various image processing applications. One image is considered to be a reference image and another a test image, the test image is transformed to the axis of reference image with the help of geometric transformation. Most of the image processing applications use integration and differentiation of data and image registration is a preliminary step for all such applications. Synthetic Aperture Radars provide wide range of data for remote sensing applications like change detection, image mosaicking, image fusion, multispectral classification etc. These applications require image registration as a primary step. The process of image registration is classified into feature based and area based. The conventional area based registration is dependent on geometric orientation of pixels and their intensities. SAR images having speckle noise may bring out failure in the registration using area based techniques. The significant features available on a SAR images aid for feature based image registration, thus making it preferable.
SIFT (Scale-Invariant Feature Transform) is a popularly used feature point detection technique for registration. To make feature detection and matching faster SURF (Speeded Up Robust Feature) is used, which is robust to rotation and scale of the images. Once the features are detected, matching of features is performed. As per the matched features the test image is transformed to get aligned with the reference image.
This thesis proposes registration of multi-temporal, multi-modal and multi-view point images using different proposed techniques. The output registered images are shown in checkered board representation. The simulation results are compared with other standard works.