Research On The Technology Related To Landmark-points Matching On Satellite Image

Landmark point is an important data for geometric correction of satellite image, theresult of landmark-point matching directly affects the geometric correction result.Therefore, the technology related to landmark-points matching on satellite image,included cloud detection, land area access and landmark-point database, is one of thekey research on geometric correction of satellite image. This paper discusses the maincontent of landmark-point database technology, and focuses on the cloud detection andland area access image technology before satellite image matching.This paper introduces the basic imaging model of the satellite sensor and the errormodel, analysis the factors affecting the accuracy of landmark point matching, includingimage noise, the deformation of image features, the point spread phenomena of opticalsystem, clouds interference and the land area in matching template. Based on themaximum correlation coefficient matching method, evaluate the factors affecting theaccuracy of landmark point matching on satellite image by simulation.To solve the problem of cloud interference and the band of satellite sensor isusually limited, a novel method based on dual-band satellite image for cloud detectionand contour extraction is proposed. At first, using an experience threshold method toidentify the cloud pixel in the dual-band images, then using the threshold segmentationmethod according to volume to take out the scattered cloud pixels or which does notshield the landmark point, and extract cloud contour based on side restrain competemethod. Using the method to process the MODIS images of different geographicalposition, the result shows this method is effective.Aiming at that direct using the image segmentation method to process satelliteimage cannot effectively get the land area, a novel satellite image land area accessmethod based on image data fusion which produced by multi-band satellite sensor ispresented. The proposed method considered with the basic theory of cloud detection andwater identification in the satellite image. At first, identify the likely land by using theradiated characteristic of the image features, and remove a part of the pseudo-land pixelusing ratio method for image data process which produced by multi-band satellitesensor. At the same time, it can expand the grayscale difference between the land andthe sea. Then using the threshold segmentation method based on the mean value to getthe possible land area. At last, to remove more pseudo-land pixel, the images producedby multi-band satellite sensor are fused by using the weighting method. Finally, a binaryimage of land area which we are interested in can be obtained. The simulation resultsshow this method can effectively get the land area.