Analysis of image inpainting and object removal methodologies

Image inpainting is a process that tries to fill in missing parts of a degraded image or remove objects from an image, but maintain the realistic content in the image. Various methods can be employed to perform inpainting. Here three benchmark methods of which two are conventional and other GAN based are evaluated to check the effectives of the inpainting process, for different class of images. The visual quality evaluation as well as quantification parameter namely the SSIM score was computed. Based on the test score it was seen that the GAN based method was capable of providing visually good restored images with an average SSIM score of 0.9516, which is an indicator on the quality of the image. The GAN based methods can be further enhanced to provide more realistic images by incorporating functions to extract both local and global contextual content.