Image Inpainting as an Evolving Topic in Image Engineering
Yu-Jin Zhang
DOI: https://doi.org/10.4018/978-1-4666-5888-2.ch122
2015-01-01
Abstract:Image inpainting (also known as image interpolation or video interpolation) is an evolving topic in image engineering (IE), especially in image processing (IP) layer (Zhang, 2009). It aims to automatically improving the quality of image, by filling in the holes of image that are to be removed for artistically or confidential considerations. Research in this field is on the rise and many applications are created (Abraham, 2012; Janarthanan, 2012; Modi, 2012). How can we make computer modify an image in a visually undetectable way analogous to sophisticated artists? This is quite an amazing topic in both art and image processing areas, with applications from the restoration of damaged paintings and photographs to the removal of selected objects. Image inpainting exactly aims to reconstitute the missing region (called the target region) using information from the remaining image areas (called the source region). Until now, there are two main categories of image inpainting approaches in the literature: diffusion-based (or PDE-based) approach (Chan, 2001; Tibshirani, 1997) and exemplar-based approach (Criminisi, 2003; Shen, 2009; Wong, 2008; Xu, 2010). In the following, two recent works, one from each category, are introduced as example (some other related methods can be found in references). One uses a patch-wise inpainting algorithm with sparse representation over a redundant dictionary. It is suitable both to deal with large holes and to preserve image details while taking less risk. To ensure the visually plausibility and consistency constraints between the filled hole and the surroundings, a redundant dictionary is constructed by directly sampling from the intact source region of current image. Then, the sparse representation for each incomplete patch at the boundary of the hole is sequentially computed and recovered until the whole hole is filled. Another uses a patch propagation-inpainting algorithm based on Weighted Sparse Non-negative Matrix Factorization (WSNMF). In this approach, the inpainting task is casted as a sequential low-rank matrix recovery and completion problem, where the incomplete data matrix consists of the image patch to be inpainted and several similar intact candidate patches under the assumption that they can be described using a low-dimensional linear model.