Publication

Abstract : There are many real world scenarios where a portion of the image is damaged or lost. Restoring such an image without prior knowledge or a reference image is a difficult task. Image inpainting is a method that focuses on reconstructing the damaged or missing portion of images based on the information available from undamaged areas of the same image. The existing methods fill the missing area from the boundary. Their performance varies while reconstructing the structure and texture present in the image and majorly fails for larger inpainting area. This paper attempts to segregate the structure and texture using Laplacian Pyramid and inpaint them separately using a top down approach. The images are inpainted from the lowest spatial resolution using Exemplar based image synthesis. The results are updated before moving to the higher resolution levels. This multi resolution process ensures the coarser details being filled before the finer details. The structure propagation is better since it is handled separately. The top down approach alleviates the traditional boundary based filling and breaks the single large sized inpainting region into many smaller sized ones as we move down the pyramid. Different types of images have been experimented and the results are summarized.