Matrix-value regression for single-image super-resolution
Yi Ping Tang, Hong Chen · 2013
Single-image super-resolution is firstly treated as a problem of matrix-value regression. By using matrix-value regression techniques, some desired properties are found. Firstly, the matrix-value regression technique greatly promotes the efficiency of learning from image pairs. As a result, the matrix-value regression based super-resolution algorithm can be smoothly applied to big data setting. Secondly, the matrix-value regression technique makes it possible to design a patch-to-patch super-resolution algorithm. As far as we know, it is the first patch-to-patch algorithm in the field of single-image super-resolution. Experimental results have shown the efficiency of the matrix-value regression based super-resolution algorithm in the training process. Meanwhile, it is also shown that the performance of the proposed algorithm is competitive to most of state-of-the-art super-resolution algorithms.