Sparse method for ringing artifact detection

The problem of non-reference ringing detection is considered. The idea of the proposed method is to decompose the input edges into the sums of blurred edge, ringing oscillations and the residual using sparse representation approach with the pre-generated dictionary. Edges with ringing effect are modeled by applying ideal low-pass filter to the step edge with different cut-off frequencies. Then they are separated into edge and ringing components by subtracting the ideal step edge blurred with Gaussian filter. The shifted and rotated blur and ringing components form the dictionary used for sparse representation. The presented method performs well on images with ringing from different source types and is robust for the images with noise.