7. Female Facial Beauty Attribute Recognition and Editing

Many computer vision researches aim at building automatic systems that can analyze a broad range of visual concepts. In this chapter we are interested in those concepts that are related to human social attributes. The chapter reports an intelligent systems that can first identify face from social media, then recognize the age, gender, and beauty score attributes. Most importantly, the system can automatically edit the facial image to get it younger and more attractive, and hence making the social media sharing experience more pleasant. We applied deep learning and multitask learning techniques to train a convolutional neural network model for the above recognition task, and applied a gradient-based method to generate the editing information that can beautify the facial image. Our work goes beyond previous works in several aspects: (1) The fully automatic system does not require costly manual annotation of landmark facial features but simply takes the raw pixels as inputs for recognition; (2) The constructed neural network model can be used to both beautify and beastify the facial image, with the amount of modification controllable by a single regularization parameter; (3) we imposed no restrictions in terms of pose, lighting, background, expression, age, and ethnicity on the images used for training and testing; and (4) we further developed a visualization method to interpret the learned model and revealed the existence of several features related to these human attributes.