Improved Digital Image Restoration Algorithm Based on Criminisi

Digital image inpainting pertains to the reconstruction of missing or damaged portions of images in a manner that is not detectable to the observer. This study evaluates the Criminisi algorithm for digital image inpainting and proposes four improvements. First, a new priority function was introduced using piecewise function to adjust the sequence of inpainting process and change the operation used from multiplication to addition under certain conditions. This adjustment improved the flexibility of the algorithm thereby avoiding incorrect filling sequence caused by rapid decay of data term. Second, Sobel operator instead of the traditional gradient direction was used to improve the computation of isophote. This approach first restored points on the isophote, which is beneficial to information fusion. Third, a new matching function was proposed and a new matching search method was employed to identify samples in the neighborhood of the damaged region based on similarity. This improvement created the nearest optimal matching block used for inpainting. Finally, a new formula of smoothing error propagation for updating confidence value was defined. This method ensured the correct sequence of synthesis from the periphery to the center. Experimental results show that this improved algorithm yields satisfactory inpainting results and improves repair efficiency. Subject Categories and Descriptors K.2.8 [Digital Image Processing]: Digital Image Inpainting; B.2.4 [Improved Arithmetic]: Algorithms General Terms Digital Image Inpainting, Algorithms