Unsharp Mask Sharpening Detection via Global Analysis

Image sharpening prediction plays a significant role in image restoration, forensics, and computer vision. This work proposes a new method for detecting sharpening and indexing in digital images. Images initially undergo manipulation using Gaussian filters and Unsharp Masking (UM) with various degrees of intensity. This is followed by transformation into the frequency domain using Fourier transform. The resulting amplitude values are calculated, and their absolute values are summed along the frequency axis to produce a profile curve. These data curves are subsequently smoothed. The UM curves exhibit concavity and symmetry relative to the curve of the original image. The values of these curves at the concave positions, corresponding to various UM intensities, can be utilized as a sharpness index by comparing them to the curve values of the original image at the same positions. These sharpness indices are dependent on the UM amplification factors. This methodology finds utility in medical image processing and computer vision systems for sharpness detection and indexing. The features identified in this study, based on Fourier power spectrum analysis, can be input into a machine learning classifier for classification.