Understanding Beauty via Deep Facial Features

Abstract

Figure 1. In each image pair, which one (Left or Right) is more attractive? We propose a method and a novel perspective of beauty understanding via deep facial features, which allows us to analyze which facial attributes contribute positively or negatively to beauty perception. To validate our result, we manipulate the facial attributes and synthesize new images. In each case, left corresponds to the original image, and right represents the synthesized one. The sample modified facial attributes from left to right are small nose to big nose, male to female, no-makeup to makeup, and young to aged. To see our discovery to the first question, please read remaining of the paper. AbstractThe concept of beauty has been debated by philosophers and psychologists for centuries, but most definitions are subjective and metaphysical, and deficit in accuracy, generality, and scalability. In this paper, we present a novel study on mining beauty semantics of facial attributes based on big data, with an attempt to objectively construct descriptions of beauty in a quantitative manner. We first deploy a deep Convolutional Neural Network (CNN) to extract facial attributes, and then investigate correlations between these features and attractiveness on two large-scale datasets labelled with beauty scores. Not only do we discover the secrets of beauty verified by statistical significance tests, our findings also align perfectly with existing psychological studies that, e.g., small nose, high cheekbones, and femininity contribute to attractiveness. We further leverage these high-level representations to original images by a genera-tive adversarial network (GAN). Beauty enhancements after synthesis are visually compelling and statistically convincing verified by a user survey of 10,000 data points.

Facial Beauty Prediction and Analysis based on Deep Convolutional Neural Network: A Review

“…Unsupervised beautification model does not depend on the truth target facial images that have been established. A novel study was proposed by Liu et al [52] using a mining beauty semantics of facial features. they relied on big data to construct descriptions of beautiful faces in quantitative and objective manners.…”

AnonymousNet: Natural Face De-Identification With Measurable Privacy

“…We are also grateful for Professor Fengqing Maggie Zhu and anonymous reviewers from the CV-COPS'19 program committee for their comments and helpful suggestions. The idea of this paper was born when the first author writing [35] during his research internship at ObEN, Inc. He thanks colleagues at ObEN for the wonderful times they shared.…”

Deep Shapely Portraits

“…It focuses on salient structure (facial features and contour proportions) adjustments instead of considering the underlying physiological structure of the human face, thus yielding artifacts. Portrait texture/shape editing methods are mostly based on generative models [21,29]. They can change both the face texture and shape of portrait images.…”

“…As a result, portrait editing plays a significant role in many applications, such as social media, advertisement, visual effects, fitness incentives, etc. In practice, the facial attractiveness of a portrait image can be affected by two main factors, face texture and face shape [24,27,29]. The former is solely determined by the colors on the face and can be improved by direct color adjustments such as brightness enhancement [28,36].…”