A face beauty prediction method based on deep self-learning

Abstract

The invention discloses a face beauty prediction method based on deep self-learning. The local texture feature of the face image extracted by the LBP operator has strong robustness to illumination and small translation, and it is helpful to use it as CDBN input. Based on the network's associative memory of the beautiful information representing the face, it further reduces the unfavorable feature description learned by the network. The self-learning method can automatically improve CDBN's understanding of the distribution of data features, so that face beauty prediction can still obtain a better description of face image features when the type and number of training samples do not meet the actual needs. By learning a deep nonlinear network structure, the network system does not rely on manual feature selection to combine low-level features to form a more abstract and structural high-level distributed beautiful feature representation, expressing an automatic learning and feature extraction process, and using The SVM regression method achieves a high degree of consistency between machine scoring and manual scoring for facial beauty.

Description

technical field [0001] The invention relates to the field of human face beauty evaluation by computer image processing, in particular to a human face beauty prediction method based on deep self-learning. Background technique [0002] It is human nature to love beauty, and everyone has the heart to love beauty. Aristotle said: "A beautiful face is a better testimonial". The favorable impression of beauty exists in daily life and has a great influence on people's daily life. The research on the beauty of human face is a frontier topic in the research on the nature and laws of human cognition that has emerged in recent years. Exploring how to better predict beauty will help the eternal theme of human beings, the code of human face beauty, to be scientific, objective and quantifiable. The description of human face beauty has made great progress in the interdisciplinary field of human face beauty research. [0003] At present, scholars at home and abroad mostly use geometric f...

AI Technical Summary

Problems solved by technology

However, the beauty of a human face largely depends on the local structure and contour information of the human face, while the mainstream methods based on appearance features only use primary features such as eigenfaces and texture features to represent facial beauty information, and do not involve more Structural, hierarchical feature expression

The traditional machine learning method for face beauty research requires a large number of training samples, and the degree of beauty requires a certain degree of discrimination, that is, the training samples should contain many labeled extremely beautiful and extremely ugly images, but in reality, most of the face images are neutral beauty. , it is difficult to find many extremely beautiful and extremely ugly images, which will undoubtedly bring difficulties to the research work

Images