Single Image Superresolution Using Maximizing Self-Similarity Prior

Figure 1

The comparison between the results of EPLL and the proposed method. (a) The ground truth image. Through downsampling and upsampling the image by the bicubic method, we derive (b), which is used as the input for producing (c) and (d) using the EPLL and the proposed method, respectively. From (c), we can see that EPLL fails to capture the image texture and its fine details. In comparison, the result of the proposed method (d) appears more natural and realistic; in particular the edges and textures regions are better preserved.

(a) Ground truth of “bear”

(b) Bicubic

(c) EPLL [23]

(d) Ours