Abstract: Embroidery on clothing patterns not only perfectly combines traditional culture with ancient techniques, but also inherits pattern patterns and traditional craftsmanship. Synthesizing embroidery images with complex textures from color images is a challenging task. Currently, stylization conversion methods based on deep learning can lead to confusion in embroidery stitch texture and color shift in the original image when rendering clothing pattern images. In order to solve this problem, an algorithm for embroidery stitch generation and modeling based on generation learning is proposed. This article starts from the material and stitches of the embroidery substrate, and uses computer graphics knowledge to establish a mathematical model of the geometric structure of the stitch primitive. According to the regularity of stitch primitives, this paper divides embroidery stitch into two categories: regular stitch and irregular stitch. Generation learning method is used to generate and simulate representative embroidery stitch in the two categories, thereby constructing a set of embroidery stitch database.

Key words: generative learning, stitch primitives, stitches reconstruction, embroidery rendering