A new framework of normalized convolution for superresolution using robust certainty
Feng Xu, Huibin Wang, Lizhong Xu, Huang Chenrong · 2010
In the process of recording a digital image, superresolution (SR) is a feasible soft method for solving the limitation of device and effect of environment. During last two decades, many researchers proposed various SR algorithms for image reconstruction. Among these algorithms, normalized convolution (NC) is a promising approach which considers not only spatial distance between center pixel and neighbor pixel but also residual error between pixel's measurement and estimation. However, the problem of removing noise and outlier in traditional NC can be further studied and solved. In this paper, we proposes a new framework of NC. Besides above factors considered, the new NC using idea of bilateral filter considers additional factor: photometric difference between neighbor pixels, so it can obtain more accurate result. On the other hand, this paper suggests a robust certainty function in NC based on two traditional certainties to detect and remove more outliers. Experiments are carried out to demonstrate the effectiveness of our method.