Abstract
Principal Component Analysis (PCA) is commonly used for facial images representation in global face super-resolution But the features extracted by PCA are holistic and difficult to have semantic interpretation For synthesizing a better super-resolution result, we introduce non-negative matrix factorization (NMF) to extract face features, and enhance semantic (non-negative) information of basis images Furthermore, for improving the quality of super-resolution facial image which has been deteriorated by strong noise, we propose a global face super resolution with contour region constraints (CRNMF), which maks use of the differences of face contour region in gray value as face similarity function Because the contours of the human face contain the structural information, this method preserves face structure similarity and reduces dependence on the pixels Experimental results show that the NMF-based face super-resolution algorithm performs better than PCA-based algorithms and the CRNMF-based face super-resolution algorithm performs better than NMF-based under the noisy situations.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have