AUTOMATIC MASK GENERATION FOR ROBUST IMAGE AND VIDEO INPAINTING APPLICATIONS

Abstract

Image Inpainting is a technique to modify an image in undetectable form filling in the missing regions in Images or `holes' using whatever additional information can be obtained from surroundings. Vast progress has been made by techniques that use the immediate boundary of the region and some additional information on images to solve this problem. In order to carry out the Inpainting operation on an image user needs to provide an appropriate mask marking the region to be inpainted and this step requires heavy manual intervention. None of the Inpainting research addresses this issue of Inpainting complexity satisfactorily. Locating the inpainting domain is a challenging task for every such application. But vastly different completions may be desired even when the local neighborhoods are identical depending on surrounding spatial and/or temporal information that may be different in these cases. In this dissertation we develop and implement a diversified framework that can be applied to various inpainting applications in a customizable manner. This framework assists in defining the mask images for certain applications taking the benefits from the application specific domains. The applications covered under this framework are scratch detection from digital media, text region segmentation from any standard font text image, and Beard Removal from human face image