Remote Sensing Image Registration, Using the Integrationof Harris Operator and SIFT Descriptor

Image registration is a fundamental technique in image processing which plays a main and critical role in photogrammetry and remote sensing. The SIFT (Scale Invariant Feature Transform) algorithm is among the most powerful matching approaches which has recently been applied successfully in fully automatic remote sensing images registration. But the feature detector in SIFT algorithm still have basic weaknesses. High computation time, and low controllability of SIFT detector are among the main challenges with this algorithm in remote sensing image registration. In this paper a new approach for remote sensing image registration is designed to reduce the computation time and increase the efficiency of standard SIFT algorithm. In this approach the SIFT detector is replaced withHarris corners in pyramidal structure. Using a specific selection strategy for the best corner features, based on stability and distinctiveness constraints, is the main characteristic of this approach which is performed in full distribution of location and scale. After feature extraction and description, initial matching process is done employing a dual-strategy using Euclidean distance in multilevel form. In the next step, based on projective transformation model and the least squares method, accuracy of initial conjugates are evaluated and then the removal of inconsistent conjugate pairs is performed.Registration approach is completed with a piecewise transformation between the reference and input images. Experimental results for four multisensor and multidate remote sensing images proved the high efficiency improvement in comparison to standard SIFT method.