Patch-based Face Hallucination with Locality Preserving Projection

In video surveillance, the resolution of face image is very important for Police Criminal Investigation. A novel method for super-resolution of face images is proposed in this paper. The method use LPP(Locality Preserving Projection) to discover the hidden nonlinear structure in high-dimensional data space, and construct the feature subspace of face image by analyze its internal manifold. We divided the input low resolution face image into patches and select the nearest patches for each patch in the subspace. Then use the selected samples restore the high-resolution face images in the absence of high frequency components. Experimental results demonstrate that the proposed method outperforms the traditional schemes significantly both in subjective and objective quality.