The requirements of light-weight and low-power of portable devices in applications involving super resolution make it necessary to design the underlying algorithms with small number of parameters. In this brief,… Click to show full abstract
The requirements of light-weight and low-power of portable devices in applications involving super resolution make it necessary to design the underlying algorithms with small number of parameters. In this brief, based on the idea of forward prediction of adaptive signal processing, a novel super block is developed for the task of image super resolution by a light-weight neural network. The design of the super block is based on using a sequence of dense residual blocks and recalibrating their outputs by a squeeze-and-excitation unit, in order to implement the idea of forward prediction. It is shown that a network that employs our super block provides a performance superior to that of the other light-weight deep networks for the task of image super resolution.
               
Click one of the above tabs to view related content.