Adaptive Manifold Learning Method for Face Hallucination

The size of training set as well as the usage thereof is an important issue of learning-based super-resolution.This work presents an adaptive learning method for face hallucination using Locality Preserving Projection(LPP).LPP is an efficient manifold learning method that can be used to analyze the local intrinsic features on the manifold of local facial areas by virtue of its ability to reveal non-linear structures hidden in the high-dimensional image space.We fulfilled the adaptive sample selection by searching out patches online in the LPP sub-space,which makes the resultant training set tailored to the testing patch,and then effectively restored the lost high-frequency components of the low-resolution face image by patched-based eigen transformation using the dynamic training set.The experimental results fully demonstrate that the proposed method can achieve good super-resolution reconstruction performance by utilizing a relative small amount of samples.