Radial Basis Function Networks for Image Restoration with Stochastic Normalizations as Bayesian Learning in Deep Conventional Neural Network

The objective of the image restoration process is to improve the high-quality visual resolution from degraded images using various techniques in Artificial Intelligence., utilizing a normalization method, a method of centers and weights that has been adapted from the Radial Basis Function Network (RBFN) is initiated to the Stochastic Normalizations method as a Bayesian Learning algorithm. The normalized form of the RBFN-SB algorithm is derived, and the RBFN input data vector is from a damaged image array raised using Taylor's expansion. Following the dynamics of the normalized technique appeared to be tactful in comparison with the RBFN-SB. The test performance comparable to BN and, concomitantly, better validation losses suitable for ensuring results unpredictability estimation through approximate Bayesian posterior are procured. To test the outcome of the proposed system gave speed up the learning process, with minimum mean squared error (MSE) and produced better results of restoration images from damaged images. In the same form, the normalization accuracy can also be improved notably by the Bayesian learning algorithm.