摘要
提出了一种图像生成的概率模型,通过检测人脸区域与该区域内人脸的特征,获得最佳的推理算法.该方法将图像分割成若干任意大小区域,包括人脸区域与背景区域,其目的是对相似度模型进行改进,以便判别人脸与背景区域的生成部分,然后利用GentleBoost算法定位出任意图像的人脸和人眼部分.实验结果表明,采用该方法能获得较好的效果,具有一定的使用价值.
In the paper, a probabilistic model of image generation was proposed and the optimum inference algorithm was derived through detecting faces and features within the face framework. The images were divided into patches of arbitrary size, containing the face and background district. The purpose of doing so was to improve likelihood-ratio models for face versus background generated patches and locate the faces and eyes on arbitrary images by Gentle Boost algorithm. The results demonstrated that this algorithm is feasible and applicable.
出处
《海南大学学报(自然科学版)》
CAS
2009年第4期395-399,共5页
Natural Science Journal of Hainan University
基金
海南省自然科学基金项目(80638)