Neural Painting based on Image Decomposition

Painting is a complex and creative process that involves the use of various drawing techniques, along with brushes dipped in paint, to create beautiful artworks. The concept of training artificial intelligence models to imitate this process is referred to as neural painting. Prior neural painting methods have focused on minimizing the difference between the output image and the original image. However, such methods often produce strokes with multiple colors, failing to accurately capture the human choice of paints during the creation process. To address this limitation, the paper presents a layer-based neural painting model. Our model decomposes the input image into several layers, each consisting of a single paint and being painted by a fixed color stroke. We propose a lightweight network that uses a series of stroke parameters as input and an alpha map as output, with colors predetermined. This method avoids the stroke parameter coupling problem that has been encountered in previous methods. The layer-based neural painting model allows for greater control over the painting process, enabling the generation of high-quality artwork with a more authentic look and feel. Experiments demonstrate that the paintings generated by our proposed method are more realistic and vivid.