Facial Super-resolution Reconstruction of Sparse Dictionaries With Posteriors Information

To improve reconstruction quality of face image under surveillance environment,the construction of robust dictionary and facial super resolution based on posterior information method were proposed.This method utilized the image dataset collected from the spot of surveillance to train a lowd-imensional dictionary,which was only sparse to the clear content part of the input image but not to the noise residue part,and a corresponding high-dimensional dictionary.Then the coefficients calculated from low-dimensional space were mapped directly to the high-dimensional space in order to synthesize the reconstructed facial image.Consequently,robustness of local facial super resolution based on sparse representation to surveillance noise was improved.Resultsdemonstrate that,for the surveillance images,the proposed dictionaries based on posterior information are more robust than the traditional schemes,and the reconstruction results obtain a better subject quality compared with the traditional dictionary.