Research On Content-based SAR Image Compression And Coding For Man-in-the-loop

In favor of the application requirement of "man-in-the-loop" image precision guidance of cruise missile, aiming at solving the problem how to transfer SAR images in a limited channel instantly and effectively, some relating key techniques such as SAR image decomposition, intelligent and fast detection of target's ROIs (region-of-interests) of SAR images, intelligent compression coding of SAR images and the quality evaluation of SAR image compression, etc, are systematically studied by utilizing nonlinear diffusion model, the theory of multi-scaled geometric analysis, and the intelligent detecting method. The main content and contributions in this dissertation are as follow.Nonlinear diffusion based SAR image decomposition is thoroughly studied in this dissertation. A diffusion parameter which represents the product of local coefficient of variation and window amplitude mean is proposed due to the defects of gradient parameter and local coefficient of variation in SAR image decomposition. This parameter can discriminate the ROI and background better. Based on the above-mentioned parameter, according to the relationship between robust estimation and nonlinear diffusion, the Turkey loss function is introduced as the diffusion function in SAR image decomposition because of its better performance, and the contour of an image is extracted by the Turkey diffusion function with the proposed diffusion parameter mentioned above. The experiment results indicate that the Turkey loss function based diffusion process can strengthen the conspicuous contour. Additionally, a Raita's (pa ta) criterion-based method, solving the automatic diffusion threshold, is proposed to automatically set the threshold in diffusion decomposition.The intelligent and fast detection method of target's ROIs for SAR images is also studied in this dissertation. A task-driven uniform framework for intelligent and fast target detection is designed. The framework has a good scalability and various when realization but has a similar general flow. Under this framework, the Parzen-window function based global CFAR detection in structure component as a coarse detection algorithm is proposed. This algorithm can provide the prior knowledge of large-scaled geometric structure for further accurate detection, also, the global CFAR detection can realize fast image prescreening; Further, based on the practicability of target detection algorithm, an intelligent CFAR algorithm for the original SAR image is proposed; Finally, in order to meet the real-time request of precision strike, a dataset splitting-merging based fast algorithm is proposed. Theoretical analysis and experiments testify the performance of the algorithm.SAR image compression coding method based on image decomposition and ROI protection combining partial differential equation (PDE). Wavelet and 2G Bandelets is thoroughly studied. A novel image decomposition and compression framework based on PDE method, 2G Bandelet, and directional interpolation and prediction is proposed. In this framework, calculating of mATRix deviation degree brings the accurate location of predicting direction, and the directional prediction and fractional interpolation inhibit the false edge of the images while keep the original edge, so these techniques improve the quality of image compression. Besides, an improved quadtree decomposition method is proposed due to weakness (such as higher computation complexity, higher memory occupation, and low efficiency, etc.) of 2G Bandelets sub-band quadtree decomposition. This new decomposition method has higher efficiency and lower memory occupation. Targets at terminal guide process of cruise missiles have quite an excessive amount and vary a lot, thus, an improved rate-distortion optimization based ROI coding algorithm is proposed. The proposed algorithm can avoid the problem of coding efficiency decreasing, which is due to introducing too many bit-planes, at the same time, this algorithm can code multiple ROIs with arbitrary target shape, and can do restoration and reconstruction quickly.Evaluation metrics of ROI detecting and SAR image compressing are also thoroughly studied. First, in order to evaluate the performance of content-based compression algorithm, an evaluation method of target's ROI detection based on the multi-hierarchal gray correlation projection model is proposed by being inspirited by the idea of information fusion and system engineering; second, after analyzing several traditional quality evaluation methods, a structural similarity based image evaluation method, SSIM is introduced. Utilizing the characteristics that the objective human vision is relatively sensitive to the explicit degree of edge, an evaluation metric called as ESSIM is presented by combining SSIM and the explicit degree of edge. ESSIM describes the edge feature more accurately, it is an advanced evaluation system of image quality.In summary, the work in this dissertation has solved several hard problems (including low compression rate, bad compression quality, and moderate coding efficiency, etc. under the condition of low band-width) of image compression coding in the data link of flighting control of cruise missiles, and provides a foundation for transferring high quality images instantly in man-in-the-loop control of the cruise missile.