Research On Image Restoration Method Based On RBF Network And Its Application

Neural network is one of important research fields in statistical machine learning.Radial basis function network is a model of neural networks.It has the characteristics of "near activation and far inhibition".It can solve some interpolation problems primely.Image restoration is an interpolation problem,which has been a research focus in recent years.The RBF network can be used for image restoration,which can discard a series of complex mathematical calculations in traditional image restoration methods.In order to get better data sets,we compare different neighbor selection methods and sampling for different damaged types.Then,a repair function is fitted by using the sampled data set and radial basis function network to fill pixels in the damaged area.Meanwhile,aiming at the damaged images with large blocks,the iterative repair method is adopted to make the restoration area smoother.In the aspect of application,the improved generalized RBF network is applied to denoise the seismic plane.According to the characteristics of large gradient of seismic surface,the activation function of network is replaced by two element Gauss function,and the pixel value is calculated by two inputs simultaneously.In view of the excessive sampling data,the sample is clustered to calculate the data center,and the gradient based method is used to calculate the extension constants of the two directions respectively.So that we can get better denoising effect and make all kinds of sediment distribution clearer.