A layered SR restoration algorithm based on hopfield neural network
Manni Duan, Xiuqing Wu, XU Shou-shi · 2006
In this paper, the layered SR algorithm which can actualize successful SR with just several frames is presented. It is also robust to large zoom factor and imprecise registration. The use of a Hopfield neural network to decide the pixel value in high-resolution image more reliably using prior information of pixel composition determined from multi-frame images by fuzzy computing was investigated. The network converges to a minimum of the energy function, defined as a goal and several constraints. For the HNN's output ranges from 0 to 1, this paper splits the 8 bit gray image to separate 8 binary images. Furthermore, the paper defines two concepts IPI(n+r)and threshold(n+r)to retain the spatial order in high bit layer. Simulation results confirmed the effectiveness of the layered SR and demonstrated its superiority to other super-resolution methods.