Detection and Removal of Cracks from Digitized Paintings and Images by User Intervention

An innovative technique for the detection and elimination of cracks in digitized paintings and images is proposed in this paper. Cracks degrade the quality of painting as well as authenticity of painting becomes questionable. In the proposed method the cracks are detected by thresholding the result of the morphological top-hat transform. Further, misidentified cracks are detected either by involving user intervention or by using a semi-automatic procedure based on region growing technique. Finally, crack interpolation also called crack filling is performed using order statistics filters so as to restore the cracked image. The true positive rate and false positive rate are used to evaluate the performance of the proposed technique. For experimental purpose we collected 2000 classified cracked and un-cracked digital paintings and images from online digital art database. The result shows achievement of true positive rate of about 98.3% at the rate of 0.1 false positive per image. This is because of providing user intervention during module called identifying mis-identified cracks.