Contextual Feature Constrained Semantic Face Completion With Paired Discriminator

Image semantic completion is to employ remaining image information restore the damaged or missing areas. Face task usually more challenging than other inpainting problems as it requires stronger consistency. We proposed a contextual feature constrained DCGAN with paired discriminator inpaint face images, which capable of overcoming DCGAN's shortages insufficient learning capability and unstable training process. Our network composed an encoder-decoder generator local global (paired) adversarial discriminator. Generator used produce parts, discriminators evaluate authenticity generated parts consistency completed respectively. In addition, generates by optimizing three types loss functions, i.e., reconstruction loss, matching losses losses. The experiments on celebA Stanford Cars dataset show that our model could generate reasonable repair results some evaluation indicators were higher baseline most test datasets.