Mask Image to Real Image Generation Based on Semantic Control Context Encoder

Abstract

In the field of image inpainting, there are some deep learning schemes, but the pixel inpainting of these schemes generally does not consider the semantics of the image. In this paper, the Semantic Control Context Encoder(SCCE) is proposed, which combines the confrontation network of text-generated images with traditional image restoration to form a comprehensive image restoration method. In this method, a context encoder is used as the generator, and a picture generated from the text is compared with the restored pictures. At the same time, the difference between the text itself and the restored picture mapped to the same space is regarded as the loss to judge the restored result, thus introducing the semantic meaning represented by the picture generated by the text on the basis of the original context encoder, and increasing the rationality of the generated picture. Experimental results on the open data set show that the proposed algorithm is superior to the traditional context encoder algorithms and the edge first algorithms.