Research on Painting Style Transfer Algorithm Based on Non-Negative Matrix Factorization

Abstract

With the ongoing evolution of deep learning algorithms, the advent of convolutional neural networks and generative adversarial networks has propelled rapid progress in the study of painting style transfer. This paper proposes a computer-assisted intelligent style transfer algorithm for modern paintings, focusing on the Non-Negative Matrix Factorization (NMF) approach. The investigation is centered on its application in modern painting style transfer, specifically addressing the challenges posed by the characteristics of stylistic and thematic content extraction in real paintings. The paper thoroughly examines both the theoretical foundations and practical implementation of the proposed method, highlighting its ability to extract authentic and significant style and content features of the give paintings. Through the utilization of NMF, the objective is to capture the global features of paintings is crucial. The edge feature extraction network can extract the edges of the original painting and input them into the network alongside the original painting. This process helps in obtaining richer detail information, leading to better perceptual results. Experimental results indicate that the proposed algorithm outperforms several existing models in terms of IS and FID indices. This suggests the superiority of the constructed model in style transfer, as it can better retain the detailed information of the original painting, leading to a more favorable subjective visual perception effect.