Image Deblurring And Dynamic Fusion With Sparse Proximal Operator

Image restoration and fusion are important to sub-system of space-based surveillance system.Moreover, it is very important to measure the relative pose, speed of the non-cooperative spacecrafti.e. pitch, yaw, roll, velocity and relative velocity and others. These parameters are essential to othersubsystems of the orbital spacecraft, which are key components for autonomous rendezvous and cap-ture system. In this dissertation, the research focuses on image deblurring and dynamic image fusion.The main contributions of this dissertation are summarized as follows:1. A sparse proximal Newton method for image deblurring is proposed. The key idea of the pre-sented method is to deal with the limitations of proximal Newton splitting scheme, which char-acterizedbyutilizingthesparsepatternoftheapproximatedpenaltyparametermatrixandrelax-ing the original assumption on the constant penalty parameter. The proposed method provides acommon update strategy for weighting matrix and a feasible solution the resulting sub-problem.Comparedwiththestate-of-artoptimizationmethodsinseveralnumericalexperiments, thepro-posed method demonstrates the performance improvement and efciency.2. An efcient optimization framework and its approximate solution for image de-blurring prob-lem, named generalized proximal conjugate gradient method, are presented. This frameworkcan be viewed as a variable hyper-ellipsoidal norm based proximal conjugate gradient method.The proposed framework may be of functional signifcance to the other large scale problems.To deal with the limitations of the proposed framework, an approximation solution is presented.The experiment results based on the pictorial and numerical results of the referred methods,indicate that the proposed method is superior to the other methods.3. A spatial-temporal dynamic image fusion algorithm, based on Kalman fltered-compressedsensing, is presented. Assuming that the dynamic fusion process in the spatial and temporal direction is separable, an estimation fusion model with a discriminative measure is designedfor modeling and combining the image signals’ space-time structures immediately. In compar-ison to the well-known multi-scale transformation-based fusion methods, this distance-learningbased fusion scheme extended from the KF-CS makes a promising improvement on objectivefusionperformanceanddeliversacompetitiveappearance. Theexperimentresultsalsohaveex-hibited that the proposed method has a remarkable dynamic fusion performance over the othermethods.