Face Hallucination Based on Partial Least Squares
Xiaohai He · ACTA PHOTONICA SINICA · 2009
A super-resolution algorithm based on Partial Least Squares (PLS) regression is proposed.PLS regression algorithm is introduced,and a super-resolution algorithm based on PLS regression is analyzed.High resolution image blocks use the high-frequency information as its feature,and low resolution image blocks use middle-frequency as its feature.The block overlapping methods are adopted to solve the block effect of the recovery.The experimental results show that the proposed algorithm performs good to IMDB face database and Yale face database.With 4 fold improvements in resolution (16 times as many pixels),the algorithm also yield good recovery results.The results of the algorithm are closer to the real images,with a higher peak signal to noise ratio.