英文摘要 | Due to the effect of atmosphere turbulence, the image or video from long-range imaging system is geometrically distorted, blurred, jittered and noised, which makes the detection, recognition, location and tracking of the target difficult. Therefore, removing the turbulence effect for high-quality target image is important. However, restoring the degraded image is challenging and one of the biggest challenges is the coexistence of dense nonrigid geometrically distortions, space-time varying blurs and sensor noise. In this work, we research the restoration of the degraded image under aero-optic effect, which has important theoretical significance and broad application prospects. We use nonrigid image registration, image denoising and image fusion to solve the challenging image restoration problem. Firstly, we couple the parametric and the non-parametric transformation model, and propose two nonrigid image registration algorithms. Secondly, we combine nonlocally sparse coding and tensor decomposition, and propose two image denoising algorithms. Finally, we unify nonrigid image registration, image denoising and image fusion and propose a restoration algorithm based on variation model and deformation information. The details and contributions are: 1.In order to correct the geometrically distortions, we propose two accurate and robust nonrigid image registration algorithms, namely, TVRC algorithm and TVL1 algorithm. Both of two algorithms combine the parametric transformation model and the nonparametric transformation model, and take the total variation as the constraint of transformation field, which make the two algorithms not only can be robust to the noise but also can align the highly-localized distortions. The difference between TVRC and TVL1 is that: TVRC uses the L2 norm to combine the two different transformation models and a two-stage strategy to optimize the objective function while TVL1 uses the L1 norm and an alternative optimizing strategy to get more accurate registration results. Experiments compared with other well-know nonrigid image registration algorithms illustrate that the proposed two algorithms can capture the local details of transformation accurately and effectively while be robust to noise. 2.In order to remove the effect of sensor noise, we propose two texture-preserved image denoising algorithms, namely, iHOSVD algorithm and NLSCTD algorithm. iHOSVD uses the nonlocally collaborative filtering technique base on Tucker tensor dec... |
修改评论