Machine learning based supper-resolution algorithm robust to registration errors

Image super‐resolution based on the pairwise dictionary selected learning and improved bilateral regularisation

“…Image super‐resolution (SR) reconstruction is considered as a process of recovering the original high‐resolution (HR) image from one or more low‐resolution (LR) images [1], which had been widely applied to video surveillance, satellite positioning, biological, and medical imaging in recent years. In general, SR methods [2–24] can be divided into three categories: interpolation‐based [2, 3] methods, reconstruction‐based methods [4–9], and learning‐based methods [10–23]. Interpolation based SR methods include bilinear interpolation, cubic convolution interpolation [2], and cubic spline interpolation [3].…”

“…Chakrabarti et al [14] proposed a kernel principal component analysis based method to produce HR face images. Kim and Kwon [15] introduced a regression based approach for producing a SR image based on the study conducted by Freeman et al In addition, Alekdandar et al [16] used machine learning to create a weight matrix for every LR image indicating the presence of registration errors. These methods made full use of the projection relation of HR image and LR image and obtained good reconstruction results but are extremely time consuming.…”

Image Super-Resolution Reconstruction Using Adaptive Co-sparse Regularization with Dual Dictionary

“…Kim and kwon [13] introduced a new method to reconstruct an SR image based on regression, the approach has been done based on study conducted by Freeman et al This method can generate good HR image while maintain the edge preserving but requires large computational time. Learning machine technique has been used to generate weighting matrix of every LR image this method can produce good SR image but consume more time to generate an SR image [14].…”