Research On Mosaic Technology For Remote Sensing Images

The steady implementation of high-scoring special projects further enhance our ability to observe Earth.Because of the limited range of single remote sensing images taken by high satellite,it is often necessary to stitch them in overlapping areas to generate a wide-angle panoramic image.High resolution satellite remote sensing images have the characteristics of high resolution and large amount of data.Although high-resolution remote sensing images are obtained,the efficiency of commonly used image mosaic methods are greatly reduced in the face of large amounts of data.And because of the chromatic aberration between the remote sensing images,it is easy to appear the obvious seams and ghosts in the image mosaic,resulting in the decrease of the visual effect of the image.For this reason,this thesis proposes an efficient image mosaic method based on CPU-GPU cooperative processing.This thesis proposes a method based on the phase correlation algorithm to calculate the image overlap range and limit the extraction and matching range of image feature points to coincident regions.This method reduces the number of feature points extracted,thus improving the overall efficiency of image mosaic.For the problem of poor efficiency and large errors in the feature matching stage of image feature points,this thesis adopts the BBF algorithm based on k-d tree for feature point matching,and uses the RANSAC algorithm to delete the bad matching pairs of feature points.This method improves the speed and accuracy of image feature point matching.A weighted average fusion algorithm is used to fuse the images.This method effectively reduces the color difference in image mosaic and weakens the visual effect of the seams.On the basis of the above CPU-side optimization,this thesis use CUDA technology to implement SIFT algorithm on GPU side.By applying the GPU to the SIFT algorithm for parallelization,combined with the operation of the CPU-side program,the speed of remote sensing image mosaic is improved.This method further improves the efficiency of image mosaic to a certain extent.In this stage,a multi-resolution fusion algorithm is used to fuse the images.This method eliminates the obvious joints and ghosts and improves the visual effect of the mosaic images.The experiments of the remote sensing image mosaic method proposed in this thesis is conducted,and compared with the existing image mosaic methods.Experimental results show that the image mosaic method proposed in this thesis can improve the mosaic efficiency of remote sensing images,enhance the visual effect of the mosaic images to a certain extent,and has a strong universality.