摘要
传统的单摄像机采集技术无法保证人脸图像的质量,直接影响人脸信息的应用价值。为此,提出一种最优人脸图像采集算法。通过估算与姿态变化具有对应关系的夹角,对基于特征点的姿态评估方法进行改进,以提高人脸姿态估计的准确率。采用多个摄像机对同一区域进行同步采集,比较同一时刻每个摄像机采集图像的肤色占有率,选出具有最佳视角的人脸图像,并全方位地判别图像中的人脸姿态变化,给出一组正面的人脸图像。在人脸图像质量评价指标中,增加人脸相似度作为图像质量参考指标,并采用一种由粗到细的评价架构。实验结果表明,该算法可提高主客观评价的一致性及人脸识别时的准确率。
The traditional single camera collection technology cannot guarantee the quality of the face image, and then directly reduces the value of face information in the practical application. Therefore, this paper proposes a best face image acquisition algorithm. In order to enhance the accuracy of pose estimation, the pose estimation algorithm is improved through evaluation angle that has corresponding relationship with face rotation. Multiple cameras are set to collect the same area synchronously. In order to select optimal view, the color skin share of image is compared with each camera gathered at the same time. All of the best perspective face images need to get through the pose estimation, and then pick out a set of frontal face images. As to enhance the coherence of subjective and objective evaluation algorithms, the accuracy of evaluation result is increased for face recognition. The method of improved image quality evaluation algorithm is put forward. Experimental result shows that the algorithm can increase face similarity as an indicator of image quality and adopt a from coarse to fine evaluation framework.
出处
《计算机工程》
CAS
CSCD
2013年第10期212-216,共5页
Computer Engineering
基金
上海博物馆视频安防系统集成基金资助项目(12-10927)
关键词
人脸图像质量评价
最优人脸图像采集
多摄像机协同
姿态估计
肤色占有率
人脸相似度
face image quality evaluation
optimum face image acquisition
multi-camera cooperation
pose estimation
complexionoccupy
face similarity
