A general residue compensation framework of learning-based face super-resolution

作者：Ma Xiang^*; Li Wenmin; Xu Hao; Yang Xiaojun; Song Huansheng

来源：Journal of Computational Information Systems, 2013, 9(10): 4049-4056.

DOI：10.12733/jcis5934

• Compensation method
• Content similarity
• Face super-resolution
• Large amounts
• Linear combinations
• Neighbor embedding
• Reconstruction method
• Super resolution

In learning-based face super-resolution algorithms, the traditional residue compensation methods are based on neighbor patches, which contain search steps with a large amount of computation and unsatisfied results. In order to improve the residue compensation step, a general residue compensation framework based on content similarity and linear combination of image patches is presented. The residue compensation framework directly uses the same content patches of the same position in the training set without searching step to compute the residue face. Different reconstruction methods can be used in the global reconstruction steps which were mentioned in this framework, thereby it has the general characteristic. Experimental results illustrate that the residue compensation method under the proposed framework can better recover the face image details, compared with the method of neighbor embedding based on residue compensation.

• 出版日期2013