Registration Of Remote Sensing Image Based On Feature Points

By using a certain similarity measurement to determine the transformationparameters, image registration is an image processing between two or several imagesunder the same scene in different conditions. With the rapid development of theremote sensing technique, as a key process of the pre-process of the remote image,image registration gets higher and higher requirements. Characteristics imageregistration is one of the common methods and its key procedure is the abstraction ofcharacteristics. Traditional image registration often uses manual or semi-automaticmethods to select characteristics. But the realization of a higher accuracy automaticimage registration is our goal in this filed. There are a lot of characteristicsabstraction algorithms and matching methods that we can use in the image registration.This study summarizes and analyzes several common characteristics abstractionalgorithms and, according to the matching problem, puts forward an improved method.And this paper mainly researches the characteristics-based image registration, whichcontains the following procedures:1. It analyzes the characteristics, merits and demerits of the characteristic points,of several operator and algorithms, including Moravec operator, Forstner operator,Harris operator, SUSAN operator. And, from the stability, noise immunity andtimeliness aspects, compares them.2. According to SUSAN operator in the shortage of rely on human’s interference,a registration algorithm of the remote sensing image based on adaptive thresholdvalue selection of SUSAN feature is proposed. The improved algorithm utilizedgradient model calculation gray histogram to get initial gray threshold, and usingvariance and OTSU in threshold correction. The threshold is optimal choice ofSUSAN feature extraction. Ensuring the gray threshold automatically is the key of thenew algorithm.3. This article studies the SIFT algorithm and the improved one. According to theproblem of time complexity and unevenly distribution of the characteristics, thisarticle improves the original algorithm and proposes a remote image matchingalgorithm, which is based on the hierarchical feature points. We make a roughmatching between images to identify the overlapped area between images. First,obtain SIFT feature points descriptors from the overlap region of base image and warp image. Second, build hierarchical feature by cutting up the image. Then select the bestmatching point using the geometric constraint of hierarchical feature to improve theimage matching accuracy. Finally, some unexpected points are eliminated byRANSAC algorithm and uniform distribution method which proposed in this paper.The experimental results show that the algorithm in this paper can realize registrationof remote sensing image accurately and effectively.