Research On Image Inversion Of Space Objects

The ground-based photoelectric detection system is one of the main technical means to detect and identify deep-space objects,and the imaging characteristic of the object is the main characteristic of using the photoelectric system to identify the space objects.Due to many factors such as imaging system,atmospheric environment and dynamic noise,the obtained space object images have serious degradation effect,which seriously affect the system's recognition and detection efficiency.Therefore,the research on inversion technology for space object degraded images has become an urgent need for space object recognition and attitude measurement.In order to obtain better image inversion effect,a sparse representation based joint sparse prior constraint blind inversion model is proposed for estimation the blur kernel.Secondly,a hybrid regularization constraint model combining the image gradient distribution prior and spatial sparse prior is constructed for realizing the inversion of space object image.This segmentation inversion method can obtain better image inversion effect.The main work of this thesis is as follows:In the process of image blind inversion,the accuracy of blur kernel estimation has a direct impact on the inversion result of the image.In this thesis,by studying and analyzing the sparse property of the space object image and the blur kernel,a joint sparse prior constraint model based on dictionary sparse representation is proposed to estimate the blur kernel.The L₀ norm of the image gradient and sparse representation is used to construct the image regular terms in the model,the robustness of the algorithm can be effectively improved while extracting the significant edge information of the image which is beneficial to the blur kernel estimation.The Laplacian distribution is used to constrain the blur kernel to ensure the sparse characteristic of the blur kernel.In this thesis,the model is iteratively solved by the alternating minimization method.In the iterative process,dynamic threshold constraints are applied to the blur kernel to ensure the continuity of the blur kernel.Simulation experiments verify that the proposed method can estimate more accurate blur kernel than several representative blind inversion algorithms.For the noisy-containing space object degraded image,the image inversion method's ability to recover the edge and texture of the image and the anti-noise ability determine the quality of the inversion image.In this thesis,by combing the dictionary sparse representation and the sparse prior information of the image gradient distribution,a hybrid regularization constraint model combining the image gradient distribution prior and spatial sparse prior is constructed for realizing the inversion of space object image.By using the global statistical characteristics of the image gradient,the model can effectively eliminate boundary artifacts caused by overlapping image blocks,and maintain the details such as edge and texture of the image,and ensure significant contrast between image pixels.Finally,the experimental results show that the proposed method has a better restoration ability for image details such as edge and texture through simulated and real degraded images.In addition,the ability to suppress noise is also better than the comparison algorithms.