The Effect of Overlap Value and Grid Coefficient on Yang Method: Super-Resolution Image Reconstruction Using Random Raw Patches

Super-resolution images are often needed in computer vision applications because having high pixel density can provide more detailed information. Super-resolution images can be obtained by interpolation-based methods and training-reconstruction based methods. Although needs more processing time, the image quality of the training-reconstruction based method is better than the interpolation-based method. This paper contributes to investigating the effect of overlap value and grid coefficient on the Yang method. This Yang method works based on random raw patches obtained through the sampling process. The overlap value and the grid coefficient are set by the user in the sampling process. In testing, we use a grayscale image with a size of 60 × 80. The image is reconstructed with magnification factor 3. The result shows that the bigger overlap value produces a smaller RMSE and a longer processing time. While grid coefficient = 1 produces a big RMSE. This is because the grid coefficient gives a few sampling points.