Abstract: This paper proposes a two-stage image inpainting model based on Generative Adversarial Networks（GANs）, which integrates improved residual blocks and a channel attention mechanism. In the first stage of model inpainting, an edge contour generator is trained to output an initial edge contour inpainting map. In the second stage, the image completion network is further employed to generate the final complete inpainted image using these repaired edge contours as prior knowledge. Experiments on the CelebA-HQ face dataset show significant improvements of this algorithm over the original and CA models, especially in key performance indicators such as mean absolute error, structural similarity index, and peak signal-to-noise ratio.