POCS Sequence Image Super-resolution Reconstruction Based On PSF Estimation And Multi-feature

In the process of forming,transmitting and recording optical remote sensing image,there are some quality degradation phenomena such as blur,distortion and noise,which greatly limit the application of image in the later period.The super-resolution reconstruction method of sequence remote sensing image has the advantages of high flexibility,effective use of image prior information,and convenient estimation of Point Spread Function(PSF).It is of great value and significance to remove or reduce the effect of noise in optical remote sensing image by using superresolution reconstruction method of sequence image,and improve the edge preserving ability and visualization effect of image.Projection onto convex sets(POCS)is a typical method for super-resolution reconstruction of sequence remote sensing images.POCS has the advantages of convenient addition of prior information,complementary information between different sequences of images,flexible imaging model,intuitive algorithm principle,and no need for a large number of training samples.In this paper,based on the principle of POCS method,POCS method is used to estimate the PSF of image by using elliptic paraboloid model,aiming at the problems of less texture details and insufficient edge preserving ability in reconstructed image.A reference frame image generation method is proposed which takes into account the local similarity feature and edge feature of image.The motion estimation method in traditional POCS method was optimized,and the sequence image was registered by Lucas-Kanade(LK)optical flow method,which improved the accuracy and efficiency of image registration.The research mainly involves the following two parts.(1)In view of the problem that the traditional POCS method does not estimate the PSF of the image,resulting in less texture details of the reconstructed image,in terms of PSF estimation,this paper adopts the optimal PSF estimation method based on the elliptic parabolic model.Based on the principle of PSF,combined with the degradation characteristics of image,the imaging characteristics of point ground objects were analyzed,and the PSF consistent with the degradation characteristics of image was obtained by model construction and parameter calculation,which provided an important theoretical basis for subsequent image reconstruction.(2)According to the implementation steps of POCS method and based on the local similarity features and edge features of images,a reference frame image generation method taking into account local similarity features and edge features is proposed to improve the edge preserving ability of reconstructed images.Based on the image edge features,the Laplacian edge enhancement method,image gradient edge enhancement method and random forest edge enhancement method are proposed.Based on the remote sensing image of "Zhuhai no.1" video satellite sequence as the data source,this paper carries out the experiment of image hyperfraction reconstruction of different ground object types,and compares and analyzes the experimental results of different hyperfraction reconstruction methods.The results show that compared with other methods,the information entropy and average gradient of the proposed method are improved by at least 0.82 dB and 0.92 dB respectively,which can effectively improve the texture detail information and edge retention ability of reconstructed images and provide better data sources for later application of images.This paper has 67 pictures,16 tables and 83 references.