Research On Technologies Of Automatic Remote Sensing Image Mosaic

Image mosaic is one of the active research areas in computer vision and image processingfield. It is the process of overlaying images of the same scene taken at different times, from thesame or different sensors. The bigger image gained after image mosaic is used for imageprocessing, analysis and application. In general, image mosaic technology includes imageregistration and image fusion. By researching and analyzing each step of image mosaictechnology, a highly intelligent image registration and automatic seam-line determinationalgorithm is developed to better realize remote sensing image mosaic. The referred algorithmis tested effectively. The main work and innovations has been done as follows:1. Based on the analysis of domestic and international research status in image mosaicprocessing, the research is focus on highly intelligent and applicable image registration andautomatic seam-line determination algorithms to improve image mosaic efficiency.2. According to the characteristic of remote sensing images and the basic principles ofgeometric imaging of the senor, three transform models involved in image registration ismentioned in detail: similar transform model, polygonal transform model and perspectivetransform. This paper evaluates the efficiency and adaptability of two local invariant featurepoints extraction algorithms: SIFT and Affine-SIFT. This paper discusses the problematicissues of the mentioned models and algorithms, and distributes Triangulated Irregular Networkmethod in image registration based on ASIFT feature points extraction algorithm. Experimentsprove that this method, compared with the common transform models, could effectivelyimprove the local accuracy of the image and assure overall accuracy after image registration.3. Through analysis and comparison of Min Gray Difference seam-line determinationalgorithm and Min Relative Difference seam-line determination algorithm, this paper proposesand accomplishes the automatic remote sensing image seam-line determination algorithmbased on extremely value estimation. Experiment proves that the determination of seam-linebased on extremely value estimation is superior to the two mentioned seam-line determinationalgorithms.4. Combined with the extremely value estimation seam-line determination algorithm, thispaper improves the seam-line removal algorithm of the pre-processing mosaic images, whichonly process pixels on one side of the seam-line and also in the overlapped region byweighting. This method better controls image ghosting and increases the image adaptation ofthe seam-line determination algorithm.