A Super-Resolution Algorithm Based on Adaptive Total Variation Regularization

An algorithm with L1 and L2 mixed norm and bilateral total variation(BTV) regularization is proposed in this paper for image super-resolution. First, the mixed norm is used as the constraint of image fidelity; Secondly, considering the effect of the BTV method is not ideal for reconstruction in the edge and texture region, an adaptive regularization parameter algorithm is proposed. In the proposed algorithm, the local structure information of image is used to control the shape and the regularization parameter, which can be adjusted adaptively according to the local structure information of the image; Finally, a minimum gradient descent algorithm is used to update the algorithm. Experimental results show that the proposed algorithm can not only reduce the mean square error, improve the peak signal to noise ratio, but also can effectively smooth Gaussian noise and salt and pepper noise, maintain the image edges and texture details.