A restoration algorithm based on parallel-channel generation convolution was proposed to address the defects of edge artifacts and semantic discontinuities when repairing large-area discontinuous semantic losses in complex images with delicate textures and backgrounds. The damaged image is input into a dual-channel convolutional structure repair network, which generates two image components with different receptive field sizes. These components are then combined through shared decoding, and the L2 loss of the output is calculated to optimize the network. Then, the output of the coarse network was sent into the fine inpainting network, which contained the residual connection and the attention mechanism to fuse context information and improve the ability to inpaint fine details. Finally, the results are sent to the SN-PatchGAN discriminator for discriminant optimization. The performance of the proposed algorithm is validated on widely recognized databases, with experimental results showing its ability to effectively restore large and irregular missing areas, even under complex backgrounds and delicate textures. The algorithm improves the authenticity and integrity of image details, structure, and features. Its peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) outperform current state-of-the-art methods.