"基于MobileNet-V2的眼镜虚拟试戴技术：解决网购眼镜试戴难题"

Abstract: The development of the Internet has greatly expanded the options for consumers when it comes to purchasing goods. Online shopping has become more prevalent, offering a convenient alternative to traditional brick-and-mortar stores. Virtual try-on technology has emerged as a solution to help consumers make informed decisions when buying products online. In today's society, the global prevalence of myopia is at an alarming rate of 22%, with China being a major afflicted country where 47% of the population suffers from myopia. As eyeglasses become a common solution for myopia, the demand for eyewear is soaring. However, certain regions may lack access to a variety of eyewear choices in physical stores, and online shoppers face challenges in trying on eyeglasses virtually, leading to misconceptions and a high rate of product returns. Utilizing virtual try-on technology can significantly reduce user biases towards product appearance and fit, thereby enhancing the online shopping experience. Virtual try-on technology is expected to prosper in the era of the Internet, promoting rapid advancements in computer vision. This paper focuses on feature point regression as the foundation, training a deep learning neural network model to regress the coordinates of the human face's eyes. By calculating the tilt angle of the eyes' coordinates, applying affine transformation to rotate images, and merging eyeglasses with the face images for dynamic virtual try-on, this technology offers a novel approach to enhancing the virtual try-on experience for eyeglasses. Keywords: Feature point regression; Virtual try-on for eyeglasses; Image merging; Affine transformation.