Style transfer in clothing based on convolutional neural networks

The combination of style transfer in the field of deep learning and clothing design has gradually become a focus of interest. Their integration can complete clothing designs and assist designers in their work. However, traditional style transfer has shortcomings such as blurred structural outlines and confused colors. This paper proposes a clothing style transfer model that incorporates structural loss and color loss. The model first converts the color space of the content image, allowing the style to be transferred onto the luminance channel of the content image, eliminating the color interference of the content image. Secondly, the Laplacian operator is introduced for sharpening both the content image and the transferred image. The sum of their variances is then added as structural loss into the total loss function, enhancing the stability of the transferred image. Finally, a linear transformation ensures that the transferred image has the same pixel mean and variance as the style image, and their sum of variances is used as the color loss in the total loss function, reducing the color differences between the transferred and style images. Experiments demonstrate that the improved style transfer model performs better in terms of texture and color than the original model.