The adaptive regularization method

The adaptive regularization method is better than the non adaptive regularization method in the case of fuzzy and noise parameter changes, which can overcome the global limitation of the regularization method. Many adaptive methods are only based on space variable smoothing operator, and the smoothing in flat region is not enough. Therefore, space variant adaptive method is widely advocated. The relationship between the complexity of regularization and the distortion rate is obtained. The adaptive recursive restoration algorithm is applied to space variant smoothing and space variant restoration respectively, and satisfactory visual and measurement results are obtained. You et al. Used the method of segmented smooth image and the adaptive restoration method based on the selection of degree and direction to smooth the spatial blur of image, protect the boundary information and reduce the ringing effect. Kang et al. Studied image restoration methods based on limited least variance (CLS). This algorithm breaks through the previous global limitation of regularization parameters, and uses every point in the signal to select regularization parameters. Due to the consideration of local statistical characteristics, the method has achieved good results. Berger et al. Proposed a space variant adaptive method. In this method, regularization and phantom suppression constraints are integrated into the algorithm at the same time, and the solution is derived by POCS. This not only improves the restoration quality, but also eliminates the ringing effect. Later experiments confirmed the validity of the method. Use optical flow to solve recovery problems. He added a parameter to the optical flow expression, which represents the reliability of the selected flow, called the reliability index. The improved algorithm can protect image edge information better. This is a relatively new way of thinking. The convergence speed is improved by improving the search direction; (2) in the preprocessing algorithm, fast adaptive iteration is realized by FIR filter structure. However, Wu et al's implementation method is different from Joon's, which is not to improve the convergence speed, but to restore the whole image directly, especially in large-scale image restoration. Although this method has not been proved in theory, it has a good effect in practice. Reeves proposes an improved fast Fourier transform (FFT) based restoration method. This method does not need cyclic convolution hypothesis and does not rely on edge information. It can not only ensure the effectiveness and real-time of the algorithm, but also suppress ringing. On the other hand, firmware or hardware can be used to implement the restoration algorithm to ensure its real-time performance. Ogrenci uses FPGA to realize the restoration algorithm, optimizes the algorithm from the hardware layer, and improves the execution efficiency.