A multilayer model of image super-resolution in the presence of inner-frame motion outliers

Accurate image registration is vital in image super-resolution. Existence of outliers, which are defined as data points with different distributional characteristics than the assumed model, will produce erroneous estimates, and then get a terrible result. Considering the outliers, Farsiu et.al. proposed a robust super-resolution method to simple reject the outliers frames but not considering the more complex motion model. This paper presents a new multilayer model of image super-resolution in the presence of inner-frame motion outliers. We first use GLOMO algorithm to separate the low-resolution image as several layers. After identify the motion models of the layers, we calculate them separately, then we can get the accurate image registration of background. At last, we fuse them into a high-resolution image. Experimental results indicate that the proposed method is better than Farsiupsilas method in the presence of inner-frame motion outliers.