Circular LBP Prior-Based Enhanced GAN for Image Style Transfer

Image style transfer (IST) has drawn broad attention recently. At present, convolutional neural network (CNN)-based methods and generative adversarial (GAN)-based have been broadly utilized in IST. However, the texture of images obtained by most presents a lower definition, which leads to insufficient details To this end, authors present new IST method based on an enhanced GAN with prior circular local binary pattern (LBP). They utilize LBP generator as improve detailed textures generated images. Meanwhile, they integrate dense connection residual block mechanism into further high-frequency feature extraction. In addition, total variation (TV) regularizer is integrated loss function smooth training results restrain noise. The qualitative quantitative experimental demonstrate that metric quality can achieve better effects proposed strategy compared other popular approaches.