Review of inpainting techniques for UAV images

Abstract

Occupying dead pixels, removing uninterested objects and shadows are often desired in the applications of an UAV to extract the natural and man-made feature boundaries. Image inpainting provides a mean to reconstruct the image. The basic idea behind inpainting methods is to naturally fill in absent or lacking portion of an image by using information from the surrounding area. Applications of this technique include the rebuilding of imperfect photographs and films, elimination of superimposed text, removal/replacement of unwanted objects, redeye correction, image coding. This paper reviews various image inpainting methods like PDE based image inpainting, wavelet-based inpainting, structural inpainting, exemplar-based image inpainting and textural inpainting with their variations. Image inpainting can also be used indirectly in squeezing image where some percentage of the original image is transmitted, and the whole image can be reconstructed on the other end using a pre-trained neural network. The critical reviews of each of these traditional methods along with the latest CNN based techniques are compared and suitability of these techniques for examining or repairing the UAV image is analyzed. In this paper, some of the existing quality assessment metrics like PSNR, MSE, ASVS, BorSal etc.related to image inpainting are also discussed.