Construction of blind restoration model for super-resolution image based on chaotic neural network

Abstract

Traditional blind restoration model for super-resolution image based on neural network is easy to fall into the problem of local minimum, so a blind restoration model for super-resolution image based on chaotic neural network is constructed. Firstly, the noise of degraded image is removed by using wavelet transform to reduce the influence of the noise of degraded image on the result of blind restoration. Then a simplified chaotic neural network model is constructed by introducing transient chaos and time-varying increment into chaotic neural network. The gray value of image is taken as the input of the network. Two Toeplitz matrices are generated by point spread function and Laplace operator. The connection weights and bias inputs of chaotic neural network are calculated by generating matrix. Iterative degradation of image gray is updated continuously according to the connection weights and bias outputs. Degree value is used to judge whether the allowable error value of network convergence meets the requirement, and when it meets the requirement, the blind restoration of super-resolution image will be output. The experimental results show that the mean values of blind restoration time, image sharpness and energy consumption are 9.273 ms, 99.045% and 118.524 J, respectively. The restoration performance of the model is good, and the blind super-resolution image has the highest similarity with the original image.