Image Style Conversion Optimization Method in Animation Design Based on Deep Convolutional Neural Networks

With the continuous development and popularization of the animation field, animation style products have received more and more attention, and the traditional style migration algorithm can no longer meet people's needs.In this paper, after researching the style principle of convolutional neural network and generative adversarial network, we propose an improved image style conversion algorithm based on convolutional neural network, introducing depth-separable convolution and cross-layer connection method to realize the information fusion of low-layer convolution and highlayer convolution, and at the same time, combining four different functions for the fitting of loss function.The proposed method in this paper is trained and validated on Facades dataset, Monet2photo dataset, Summer2winter dataset and Apple2orange dataset, and the quantitative experimental results show that the MOS value of the proposed method in this paper is improved on average compared with CycleGAN and StarGAN by about 12.67% and 52.93%, thus it can be concluded that the method proposed in this paper effectively improves the overall texture details and style quality of image style transformation in animation design.