A Progressive Image Inpainting Algorithm Based on Gated Convolution Optimization
Min Qi Zhou, Xiuyou Wang, Huaming Liu, Xuehui Bi · 2023
Image inpainting has very important practical significance and application value. Aiming at the problems of blurring or structural confusion in image inpainting for large irregular defects, we propose a progressive image inpainting algorithm based on gated convolution optimization. The algorithm accomplishes the initial prediction of image structure and semantics by adding a coarse repair network. The coarse repair network uses gated convolution to compensate for the loss of information continuity caused by dilated convolution, and introduces semantic feature loss to improve the semantic correctness of the initial repair. The refinement repair network uses recurrent feature reasoning module and applies a attention mechanism to make full use of the valid information in known regions of the image to achieve progressive image inpainting. This model is effective in restoring broken images, and the obtained image inpainting results are clearer in details. We evaluate our model on the publicly available datasets CelebA and Paris Street View to demonstrate the effectiveness of our approach through qualitative and quantitative experimental validation. The restoration results of the model show higher consistency in visual effects.