Neural network-based image segmentation for image interpolation

A novel image interpolation scheme is presented, in which a neural network is used to segment the image according to the presence of oriented edges; a set of different directional adaptive filters then interpolate the image, the filter outputs being weighted according to the neural network output. The filters are designed in order to accurately reproduce both smooth areas and sharp edges. In the paper, the structure is presented and the neural network and the directional filters are described. Simulation results show that both objective and subjective image quality obtained by the proposed method are higher than using linear interpolators.