Investigation of Exemplar based Techniques for Image Inpainting

Image inpainting is a technique for editing and estimating damaged pixels or missing part of a deformed image. This is realized by procuring nearby pixel information of the missing part or using information from external data. Various algorithms exist for restoring missing part of an image. These methods adopt either diffusion-based approach or patch-based approaches for the restoration process. Diffusion based approach use surrounding pixels information and Partial Differential Equations (PDE) to fill damaged or the missing part. This method is best suited for recovering missing image parts that are significantly small. However, in case the missing part is large then the same may be prone to visual impairments such as blurring in the restored part of an image. This is especially the case when the restored part is highly textured. Additionally, these approaches have a disadvantage of increased time complexity making it unsuitable for large area image filling. In this article we focus on the second approach viz, patch-based or exemplar-based method which is known to perform well for larger restoration areas. Experimentations are carried out to investigate the performance of this approach for different images along with parameters such as sizes of restoration patches and priority method. Quantitative results indicate that with proper choice of patches and related parameters, it is possible to obtain acceptable restoration performance even for large sized restoration area.