Image inpainting with contextual attention and partial convolution

Abstract

Image inpainting is a skill developed to recover the damage part of an images. Many researches states that application wise there are various plausible techniques implemented for an image inpainting. By using artificial intelligence here image inpainting is executed. Image inpainting utilizes the contextual attention to fill pixel wise missing region of an image, missing region filling is done by deeply understanding of available known context. Partial convolution plays an important role to reconstruct an image for matching with the realistic images with structures and textures. The work starts with applying various masks on the image and then reconstruction of an image form by applying partial convolution layer. The main difference of this technique cleared from the traditional methods of image inpainting that it is not only reconstruct the image with new content but also specifies and compress structures available in surrounding region whereas other techniques only focuses and borrow textures from the remaining image. This model is trained with two dataset where fully feedforward partial convolution network is experimented. This gives better qualitative image reconstruction because it comprises both the contextual attention and partial convolution method.