Exploring Color Styling in Landscape Design through Convolutional Neural Network-Based Image Style Transfer

In landscape design, designers typically colorize sketches according to different style preferences after completing them, which is often time-consuming and labor-intensive with high trial and error costs. Therefore, we propose an automatic image style transfer model based on deep learning to automatically colorize landscape design sketches in different styles. This model utilizes the classical convolutional neural network (CNN) VGG to extract features from input content and style images, where different network layers capture information at various levels, including texture, color (style feature), and semantics (content feature). By defining content loss and style loss to measure the differences between the generated image and the content/style images, and minimizing the weighted sum of these losses to optimize the generated image. Ultimately, achieving coloring and style generation for landscape design sketches. The experiments demonstrate that the model performs exceptionally well in coloring grayscale landscape design sketches and transforming the color style of hand-drawn landscape design sketches. The model with VGG-16 as the backbone network achieved an average loss of 13.83, while the model with VGG-19 showed a slight performance improvement, with an average loss of 8.12. The final results validate the effectiveness of the model in automatically generating styles for landscape design sketches.