Super-resolution (SR) image reconstruction has developed into a powerful tool to enhance the image resolution for the systems with low-cost imaging sensors.A direct but efficient approach to super-resolve a low-resolution image is based on prior knowledge learning.But the existing methods do not consider matched high-level features in the images with structured edges

resulting in some smooth image artifacts.A second-order directional total generalized variation (DTGV) regularization based method is proposed to explore the underlying high-level information of the data in this paper.More specifically

second-order DTGV acts as not only an additional prior but also an effective constraint to reduce the image artifacts and remove the noise.Results from several texture images demonstrate that the proposed approach can generate high-resolution image details and tend to produce high-frequency textures.