基于分形和肤色模型的自然态人脸检测方法研究

Study of face detection in natural state based on fractal and complexion model

在对当前人脸检测问题研究过程中常用方法及其特点进行论述基础上,分析了利用分形特征实现人脸检测的可行性,提出了基于分形和肤色模型的自然态人脸检测方法。通过大量实验可知,该算法既克服了传统的基于启发式(知识)模型的方法和基于统计模型方法适应变化能力差、检测速度慢的缺点,又克服了单纯的肤色模型算法对肤色相近区域误检率高的缺点,利用人脸图像所特有分形特征,可以实现自然态环境下,多角度、多姿态人脸检测。

The traditional face detection methods and their properties are discussed. Then, the feasibility of the face detection method is analyzed by utilizing fractal characteristics of human face. Finally, the face detection arithmetic based on fractal and complexion model in natural state is presented. Grounded in quantitative tests, the arithmetic can overcome the shortcomings of the low capability in adapting to changes and the low detection speed in traditional methods based on knowledge model and statistical model, and the high misdetection efficiency in the similar skin areas in the simplex complexion model arithmetic. And by using the fractal characteristics of human face, multi-view and multi-pose face detection in natural state is implemented more accurately and quickly.