Example-Based Super-Resolution via Structure Analysis of Patches

Abstract

In example-based super-resolution, it is difficult to determine appropriate high-frequency (HF) patches from a training database by using only the information of one input image. In this letter, we utilize the sharpness of high-resolution (HR) patch candidates for the reliable determination of HF patches. For each input patch, we first preselect a sufficient number of HF patch candidates and produce HR patches by adding the candidates to the input patch. After removing the outlier patches, we then reselect several HF patches according to the patch characteristic for producing the final HR image. This reselection procedure is optimized for edge patches and non-edge patches, respectively. Experimental results show that the proposed algorithm provides sharper details compared to the existing algorithms.

Super resolution reconstruction of image gradient profile sharpness

“…In most cases, this method quantitatively and qualitatively outperforms state -of the art and bicubic interpolation learning based SR approaches. Changhyun Kim et al [2] presented an example-based super-resolution, it is difficult to determine appropriate highfrequency (HF) patches from database by using only the information of one input image. In this letter, they utilize the sharpness of high-resolution patch candidate for the reliable determination of HF patches.…”

A multi-frame super-resolution method based on the variable-exponent nonlinear diffusion regularizer

“…In this paper, we used the four-point neighborhood explicit scheme to implement (12), as detailed in the later sections. Figure 3 shows the block diagram that implements our model in (12). From the illustration, a sequence of M frames, y 1 , y 2 , .…”

An integrated framework for both compression noise reduction and super-resolution

“…The integrated framework of the two algorithms is then proposed in section III and some experimental results are provided in section IV. Finally, we conclude the paper in section V. and (c) the consecutive processing of two algorithms in [8] and [2].…”

“…For the integration, we effectively combine existing patch-based SR [2] and NR algorithms [8] through the combination process of 3-D transform coefficients.…”