In this paper some new interpolation methods using cellular neural networks (CNN) are proposed. The templates have only linear feedback component with a 3*3 dimension. The templates were computed by using analytic methods. By implementing these interpolation methods on a “CNN- Universal Chip”, the processing becomes completely parallel and it can be done in real time. The testing results achieved by using these interpolation methods are presented for increasing gray- scale image resolution, for image reconstruction and the encoding-decoding the images necessary for their transmission.