Abstract: A video super-resolution algorithm based on spatial-temporal feature and neural network(STCNN) was proposed to improve the video visual resolution quality and details clarity. This algorithm comprehensively utilizes the correlation mapping relationship among external correlative blocks and the non-local similarity existed in the spatial-temporal neighboring internal blocks, thereafter, reconstructs the video details efficiently with the fitting parameters learned by deep convolutional neural network. Spatial-temporal feature similarities are introduced to optimize the reconstruction results, and to resolve the miss-match problem and improve the super-resolution performance by making full use of the complementary and redundant relationship between low-resolution video frames. Experiments demonstrate that the proposed algorithm outperforms existing algorithms in terms of both subjective visual effects and objective evaluation index.

Key words: super-resolution reconstruction, deep convolution neural network, spatial-temporal feature, non-local similarity