Fast textile pattern generation combining MRF-based and Gram-based methods

Textile pattern design is a tedious and challenging task for designers. This paper proposes a fast textile pattern generation algorithm that combines MRF-based and Gram-based methods. First, the reconstruction method based on image optimisation is determined after analysing the specific requirements of textile pattern design. The pre-trained VGG19 is selected as the style feature extraction neural network. Then, we compare the generation results of various combinations of style loss functions and propose a multi-resolution image optimisation method. Finally, the smoothing loss and colour histogram matching are added to improve the generation quality further, thus constructing an image generation algorithm for textile pattern design. Experimental results demonstrate that our algorithm can effectively generate complex textile patterns with global style and local detail features. The average image generation time is 575s, over 84.3% faster than traditional algorithms. At the same time, this algorithm is convenient for switching styles and requires lower computational capability. It can improve pattern design efficiency and promote the application of image generation technology in textile design.