Face hallucination through ensemble learning

摘要: A learning-based face hallucination system is proposed, in which given a low-resolution facial image, corresponding high-resolution image automatically obtained. This study proposes an ensemble of feature representations, including various local patch- or block-based one-dimensional vector representation, two-dimensional matrix and global representation. For each regression function constructed to synthesize from the input image. The synthesis process conducted layer-by-layer fashion, where layer composes several functions. output one then served as following layer. experimental results show that proposed framework capable synthesizing images with wide variety poses, geometry misalignments expressions even when such are not included within original training dataset.