Research And Implementation Of Large Scale Exemplar-based Image Inpainting Technology

Image inpainting makes use of undamaged information on images to restore the specified damaged regions. Applications of image inpainting include: reverting deterioration (e.g., cracks in photographs or scratches and dust spots in film), removing specific objects (e.g., removal of stamped date or sensitive information on photographs for military base), as well as special effects of film and television productions and virtual reality.At present, the techniques for the image inpainting are divided into two categories: one is based on the geometric models which suits for small-scale damage, another is based on the texture synthesis which has better results for filling large missing pieces on images. In essence, the first type is based on the partial differential equation (PDE). Typical methods are BSCB model and CDD model. Based on Markov Random Fields Model, the most notable one among the texture synthesis, Criminisi put forward the exemplar-based inpainting method which has a good inpainting resultOverall, there are some shortcomings of these algorithms. The results of inpainting with PDE algorithm are often fuzzy and it is more obvious when repairing large-scale damaged regions. On the other hand, texture synthesis method is often very time-consuming when comparing similar points. To overcome these shortages, we improve three parts as follows:1. A new algorithm is proposed for priority calculation. The calculation fuction is more reasonable and reliable to ensure the right filling order which is the most important thing to the exemplar-based inpainting.2. Narrow the scope of the search region, improve the passive and global searching for active and local searching. This small improvement makes a big progress on both the efficiency and accuracy.3. A window size adaptive algorithm is proposed. It changes the window size of the exemplar patch from fixed into alterable to further improve the inpainting efficiencyIn order to verify the validity of the algorithm, we implement the algorithm and a large number of simulation experiments have been made. After comparison, the algorithm was found not only achieved the desired effect of inpainting, but also significantly improved the speed of repair.