In addition, we also classify and summarize the different tasks of image inpainting according to the application scenarios of image inpainting. First, we review the specific research status of deep learning technology in the field of image inpainting in the past 15 years then, We deeply study and analyze the existing image restoration methods based on different neural network structures and their information fusion methods. In this article, we systematically summarize and analyze the literature on image inpainting based on deep learning. At present, image inpainting based on deep learning becomes a research hotspot in computer vision. With the rapid development of deep learning (DL) in recent years, the advantages of DL in image processing have become increasingly prominent, it can solve the problems existing in traditional image inpainting algorithms to a certain extent. Traditional image inpainting algorithms are often difficult to deal with large-scale image deletion, since these algorithms are prone to inconsistent image semantics. Image inpainting is an important research direction in the study of computer vision, and is widely used in image editing and photo inpainting etc. The experimental results demonstrate that the proposed FRRN framework for image inpainting is much better than previous methods both quantitatively and qualitatively. Correspondingly, a step loss function is applied to improve the performance of intermediate restorations. Additionally, to guarantee completion quality during progressive inpainting, we adopt N Blocks, One Dilation strategy, which assigns several residual blocks for one dilation step.
We show that well-designed residual architecture facilitates feature integration and texture prediction.
In this paper, we propose full-resolution residual network (FRRN) to fill irregular holes, which is proved to be effective for progressive image inpainting. A progressive inpainting approach would be advantageous for eliminating central blurriness, i.e., restoring well and then updating masks. However, they fail to generate plausible textures inside damaged area because there lacks surrounding information. Recently, learning-based algorithms for image inpainting achieve remarkable progress dealing with squared or irregular holes.
0 kommentar(er)
