摘要
针对快速鲁棒特征(SURF)算法冗余信息多且计算速度慢的缺点,该文对SURF算法进行改进,用于人脸检测。计算特征点邻域的图像熵,并使用非极大值抑制提取图像熵高的特征点,通过减少区域描述减少冗余信息。使用扇形窗口遍历各特征点邻域,通过计算窗口内哈尔小波响应构建该点的特征描述子,从而加快计算速度。使用费舍尔矢量核将各特征描述子映射到高维空间进行人脸检测。仿真实验表明,在应用于人脸检测数据集和基准(FDDB)数据集时,与SURF算法相比,该文算法检测率提高了7.9%,特征计算时间减少了53.1%,特征点数减少了59.7%。
An improved speeded up robust features(SURF) algorithm is used in face detection aming at the shortcomings of many redundant information and low computing speed of the SURF algorithm.The image entropy of neighborhood of each feature point is calculated,and the feature points with high image entropy are selected by non-maximum suppression to decrease description area and redundant information.A fan window is used to traverse the neighborhood of each feature point,and Haar wavelet response in the fan window is calculated to form the descriptor of each feature point,and the computing speed is faster.Each descriptor is mapped into a high dimensional space by Fisher vector kernel for face detection.The simulation results in face detection data set and benchmark(FD-DB) show that,compared with that of SURF algorithm,the detection rate of the improved SURF algorithm increases by 7.9%,the feature calculation time decreases by 53.1%,and the feature points decreases by 59.7%.
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2017年第6期714-719,共6页
Journal of Nanjing University of Science and Technology
关键词
快速鲁棒特征
人脸检测
图像熵
非极大值抑制
哈尔小波响应
费舍尔矢量核
speeded up robust features
face detection
image entropy
non-maximum suppression
Haar wavelet response
Fisher vector kernel
