Adaptive Blurring Estimation for Learning-Based Super Resolution
Yufan Chen, K. Taniguchi, Xingshuo Han · Advances in computer science research · 2015
In this paper, we address the problem of generating high-resolution (HR) image from a single low-resolution (LR) image, which is called image super-resolution (S R).Recently learning-based S R with sparse coding (SC), locality-constraint linear coding (LLC) and so on has been explored, and achieve acceptable performance.Howe ver, the conventional learning based methods cannot directly deal with a blurred LR input, which is usually considered as another research line of deblurring, and extremely difficult to imple ment for real applications.This paper proposes to firstly estimate the blurring degree of an input image, and then generate the adaptive codebook (dictionary) for learning-based S R, which can simultaneously achieve the de-blurring and high-resolution image in the learning framework.We integrate our previous LLC based S R with the adaptive de-blurring procedure.Experimental results show that our proposed strategy can reconstruct the HR images more accurately than conventional methods, and its processing time is much faster.