Abstract:  In order to express the high frequency minutia information of the image in detail, and improve the
quality of reconstructed image, a super resolution reconstruction algorithm is applied based on multi dictionary
L1/2 regularization. In the dictionary training phase, in order to effectively extract the information of feature
detail of edge and texture of low resolution image, the modified first or second order method is used to extract
feature for low resolution image. In the stage of image reconstruction, because of the problems that the solution
based on L1 regular model is usually not sparse enough and the quality of the reconstructed image needs to be
further improved, L1/2 norm is employed to substitute L1 norm to establish the super resolution reconstruction
model. The experiment shows that the present algorithm compared with the existing algorithms can better express
the section information of the image details and improve the quality of image reconstruction.

Key words:  super-resolution reconstruction, feature extraction, L1/2 regularization