人脸关键点定位中的机器学习方法
首发时间:2021-03-30
摘要:人脸作为一种信息载体具有其不可比拟的优势和重要性,其中人脸关键点定位是人脸信息分析流程中承上启下的中间环节。自上世纪九十年代以来,机器学习不断发展,对人脸关键点定位的研究乘风而起,发展至今的研究成果已经比较成熟。人脸兼具刚性和柔性双重属性,这使得早期人脸关键点定位的解决办法都是采用轮廓模型和局部特征相结合的方式。但随着深度学习的发展和大数据时代的到来,基于回归的方法框架逐渐占据主导地位。近年来,基于卷积神经网络的人脸关键点定位方法风头正盛。从人脸关键点定位研究方法的变化历程可以窥见机器学习的发展痕迹。本文综合地研究了基于机器学习的人脸关键点定位的不同实现方法,并通过理论比较分析各种框架的优点和缺点,其中ASM和AAM类的传统算法的模型具有很好的解释性,而基于回归的算法能达到更好的定位精度与速度。
For information in English, please click here
Machine learning methods for face keypoints location
Abstract:As an information carrier, human face has its incomparable advantages and important to researches in the field of bioinformation technology. And face key point location is the middle link in the process of face information analysis. Since the 1990s, machine learning has been developing continuously. The research on the location of the key points of human face has been carried out by the wind, and the research results have been more mature up to now. Human face has both rigid and flexible attributes, which makes the solutions of early face key point localization are to combine contour model and local features. However, with the development of deep learning and the arrival of Age of Big Data, the frameworks based on regression gradually have occupied the leading position. In recent years, face key point localization methods based on convolution neural network is gaining popularity. The development of machine learning can be seen from the changing of the methods of face key point localization.In this paper, different methods of face key point location based on machine learning are comprehensively studied, and the advantages and disadvantages of various frameworks are compared and analyzed theoretically. The models of traditional algorithms of ASM and AAM classes are well interpreted. The regression-based algorithm can achieve better positioning accuracy and speed.
Keywords: machinelearning facekeypointslocation CNN
引用
No.****
同行评议
共计0人参与
勘误表
人脸关键点定位中的机器学习方法
评论
全部评论0/1000