New feature selection for neighbor embedding based super-resolution

Neighbor embedding based super-resolution uses a manifold learning based on local linear embedding to estimate a high-resolution image from an input low-resolution image and a training image set. A novel feature selection combing norm luminance and stationary wavelet transform coefficients for neighbor embedding based super-resolution (NLSC-NE) is proposed. The norm luminance represents the low-frequency information or global structure, while the SWT coefficients carry high-frequency information of luminance value variations. Experiments show that compared with several existing feature selection methods, the new feature combination can capture more details and preserve edges better. The proposed algorithm improves the super-resolution performance both in subjective and objective assessments.