DATA-DRIVEN 3D EFFECT ENHANCEMENT MECHANISM OF WATERCOLOR: A NEURAL NETWORK-BASED APPROACH

Abstract

A watercolor is an art form that uses water to express the painting process. Water is the specific boundary that distinguishes watercolor from other painting types. Fluidity is one of the essential factors constituting its distinctive artistic beauty. With vibrant hues and delicate brushstrokes, 3D watercolors enhance the image’s content with more meaningful information. It has more visual impact than a watercolor that is flat. Appreciators can visualize the scenario in their minds, which leaves more room for imagination and is particularly useful for usage in animation. This paper uses 3D point cloud reconstruction to enhance the 3D effect of watercolor and proposes a 3D reconstruction method of the point cloud based on a generative adversarial network. First, a new mesh model is obtained by using a predictor to predict the offset of the mesh model edge. Then, the point cloud classifier in the discriminator is used to extract the high dimensional features of the original point cloud data and the surface sampling point set of the mesh model. Finally, the output data of the predictor and discriminator are associated with the adversarial training method, and the network model is optimized through several iterations to obtain a 3D mesh model satisfying the spatial characteristics of the point cloud. The experimental results demonstrate that the method suggested in this paper is superior to the three benchmarks in terms of watercolor’s texture, color transition, highlight part, and iteration time, as well as its ability to enhance the 3D effect.