• Using Lyapunov stability theory and artificial neural networks to create a facial beauty evaluation model
  • Ali Mohammadi Ruzbahani,1,*
    1. Semnan University


  • Introduction: In this research, using the images available in social networks, we classified only the faces in terms of similarity using image processing and classified them into 30 classes. Then we invited 230 people to rate these faces on the web application. In this software, we established each person's 30 images; it related each image to a class of the relevant classification, and people gave each image a score from 1 to 10. Images were re-classified using the received scores and Lyapunov stability theory, and users were asked to participate in the survey again. The results of the first and second surveys were entered into the artificial neural network and the neural network was able to classify the faces in terms of beauty with 95.8% accuracy. The created model can be evaluated before cosmetic surgery.
  • Methods: To analyze the images of faces, we used the OpenCV image processing library, and we also used parameters such as facial redness, the width of lips and mouth, and the length ratios of the facial parts as biasing of the artificial neural network. Then our neural network analyzed the statistical data collected from the surveys with the Reinforcement Learning method and Lyapunov’s stability function was used to adjust the neural network.
  • Results: Using the neural network trained using the results of the first survey, we evaluated the dataset related to the images of faces and compared the results with the second survey, the success of the neural network was 66.4%, again the neural network using the theory Lyapunov’s stability has been toned down and beauty criteria such as facial symmetry, dividing the width of the face into five parts and comparing the width of each part with the width of the eyes, comparing the ratio of the eyebrow distance to the hairline and the distance between the eyebrow and the end of the snout as well as the distance between the bottom of the nose and Chin is added as a bias parameter to the artificial neural network. This time, the artificial neural network was able to classify faces in terms of beauty with 95.8% accuracy compared to the survey.
  • Conclusion: By using Lyapunov stability theory and simple parameters, we can build an artificial neural network that can rate the beauty of people’s faces and use it to help cosmetic surgeons in planning and designing facial cosmetic surgeries.
  • Keywords: Artificial neural network, facial beauty, Lyapunov stability theory, cosmetic surgery, facial beauty