Face Attribute Recognition Based On Label Distribution Learning

Face attribute recognition is a key issue in pattern recognition and computer vision, including age esti-mation, gender analysis, head pose estimation, facial expression recognition and so on. It is wildly applied to different areas, such as human-computer interaction, security system, search application and intelligent trans-portation. To address problems in their practical applications, this paper proposes label distribution learning algorithm and use it to solve the head pose estimation problem and facial expression recognition problem.The main researches of this thesis include:1. This paper applies label distribution learning algorithm into emotion analysis in facial expression recognition problem. It considers that expression is mixed by different basic emotions with different intensities, other than a single emotion. We aim to predict all the appeared emotions together with their intensities.2. This paper applies label distribution learning algorithm into head pose estimation problem and predicts two angles in pitch and yaw at the same time.3. This paper solves the problems appeared in the predicted distribution.The main contributions of this paper:1. This paper firstly consider all the elements in one expression and their specific intensities as target to predict.2. We take the relationship between different emotions into consider during the learning process; 3. We add some regularization terms into the target function to solve some problems in predicted distributions, like noise and too smooth output.This paper consists of five chapters. The first chapter introduces the applications, technique components, research and challenges of the face attribute recognition. The second chapter introduces the procedure of face attribute recognition, and two feature extract methods which will be used in this paper. Besides, distribution learning method and its frame and applications also be introduced. The third chapter introduces facial ex-pression recognition based on label distribution learning algorithm. The forth chapter introduces head pose estimation based on label distribution learning algorithm. The last chapter is the summary and future outline of our work.