A Monotonically Increasing SSDA Based On Feature Of Remote Sensing Image

Image registration is the process of matching two or more images that get from the samescene derived from different time, different sensors or different views of angle. In order toimprove the system matching accuracy, reducing the matching error and adverse factors suchas noise or distortion. Before image matching images must be properly pretreated. In thispaper, the first step is image enhancement, highlighting the characteristics and enhancingcontrast; the second step is image segmentation, processing sub image, to achieve the purposeof reduce the difficulty about image matching.Image matching methods cover two categories: one is based on gray, the other is feature.The former is based on gray value of matching image. Algorithm is simple and has a highprecision, but also has the disadvantages about large amount of calculation; the latter has asmall amount of calculation, not easily affected by noise. Feature matching is not easy toinfluenced by image illumination, deformation, shelter. But the stand or fall of matchingquality often depends on the feature extraction.In this article four classic operator Moravec, Harris, SUSAN, SIFT are described. Resultof the experiments proves SIFT algorithm is invariant to rotation, scale and illumination of theirnage, and is also robust in view angle change, affine changes or noises disturbance.In this paper, we proposed two improvements based on the traditional matchingalgorithms. Firstly, get lower dimension and optimization calculation through PCA-circularstructure SIFT algorithm; secondly, reducing the amount of matching calculation based on theremote sensing image’s characteristic information. Two kinds of matching methods combinedwith an efficient way to avoid the limitations of single method. The simulation results showthat for a variety of conditions change, compared to other methods, the improved algorithmcan greatly improved the speed of registration and higher accuracy.