Image inpainting using frequency-domain priors

In this paper, we present a novel image inpainting technique using frequency domain information. Prior works on predict the missing pixels by training neural networks only spatial However, these methods still struggle to reconstruct high-frequency details for real complex scenes, leading discrepancy in color, boundary artifacts, distorted patterns, and blurry textures. To alleviate problems, investigate if it is possible obtain better performance information (Discrete Fourier Transform) along with end, propose frequency-based deconvolution module that enables network learn global context while selectively reconstructing components. We evaluate our proposed method publicly available datasets CelebA, Paris Streetview, DTD texture dataset, show outperforms current state-of-the-art techniques both qualitatively quantitatively.