The current image completion method can complete the missing image well, but they often fail to achieve better results when the missing area is at the edge of the image. In order to overcome this defect, this paper proposes a novel image completion method considering content and sharpness based on a generation adversarial network (GAN) and region growing algorithm. This model uses trained content and pixel joint discriminator to distinguish between real and generated images. The content discriminator focus on the entire image content to assess whether it is as coherent as a whole, and the pixel discriminator focus on the sharpness of the entire image to assess whether it is as same as the original image in sharpness based on multi-focus image fusion. A variety of missing blocks can be completed well by our method in dataset of Places2. In addition, compared to the other image completion method, this method can complete edge-loss blocks more excellently.