Global consistency, local sparsity and pixel correlation: A unified framework for face hallucination

作者：

Highlights：

• We improve our previous method to produce an initial high-resolution (HR) image.

• We propose a unified framework that contains the global and local priors together.

• We devise the local sparsity model, which recovers local details in patch-wise.

• The pixel correlation model further compensates the local structure in pixel-wise.

• With the initial HR image, we generate the final HR image by an iterative process.

摘要

Highlights•We improve our previous method to produce an initial high-resolution (HR) image.•We propose a unified framework that contains the global and local priors together.•We devise the local sparsity model, which recovers local details in patch-wise.•The pixel correlation model further compensates the local structure in pixel-wise.•With the initial HR image, we generate the final HR image by an iterative process.

论文关键词：Face hallucination,Regularization framework,Sparse representation,PCA position dictionary,Pixel correlation

论文评审过程：Received 23 March 2013, Revised 21 January 2014, Accepted 19 April 2014, Available online 4 May 2014.

论文官网地址：https://doi.org/10.1016/j.patcog.2014.04.023