Current Issue Cover


摘 要
摘 要:目的 三维人脸点云的局部遮挡是影响三维人脸识别精度的一个重要因素。为克服局部遮挡对三维人脸识别的影响,提出一种基于径向线和局部特征的三维人脸识别方法。方法 首先为了充分利用径向线的邻域信息,提出用一组局部特征来表示径向线;其次对于点云稀疏引起的采样点不均匀,提出将部分相邻局部区域合并以减小采样不均匀的影响;然后,利用径向线的邻域信息构造代价函数,进而构造相应径向线间的相似向量。最后,利用相似向量来进行径向线匹配,从而完成三维人脸识别。结果 在FRGC v2.0数据库上进行不同局部特征识别率的测试实验,选取的局部特征Rank-1识别率达到了95.2%,高于其他局部特征的识别率;在Bosphorus数据库上进行不同算法局部遮挡下的人脸识别实验,Rank-1识别率达到了最高的92.0%;进一步在Bosphorus数据库上进行不同算法的时间复杂度对比实验,耗费的时间为最短的8.17s。该算法在准确率和耗时方面均取得了最好的效果。结论 基于径向线和局部特征的三维人脸方法能有效提取径向线周围的局部信息;局部特征的代价函数生成的相似向量有效减小了局部遮挡带来的影响。实验结果表明本文提出的算法具有较高的精度和较短的耗时,同时对人脸的局部遮挡具有一定的鲁棒性。该算法适用于局部遮挡下的三维人脸识别,但是对于鼻尖部分被遮挡的人脸,无法进行识别。
3D Face recognition under partial occlusions based on radial strings


Abstract: Objective Face recognition technology has extensive applications in many fields because of its friendliness, nature and intuitive. According to input data, face recognition can be divided into 2D face recognition and 3D face recognition. The traditional 2D face recognition technology is based on image information or video information, although it has achieved a great success, it is still very difficult to solve the problems that caused by illumination, posture and makeup. Compared with the traditional 2D face recognition, 3D face recognition is based on the 3D data of human face, such as 3D point cloud and 3D mesh. On the one hand, 3D face recognition technology is less affected by illumination, pose and makeup. On the other hand, partial occlusion in 3D face is an important factor that affects the accuracy rate of 3D face recognition. However, the collected face data is always occluded by external objects, such as hands, hair and glasses. Therefore, 3D face recognition with partial occlusions becomes a very important research subject. In order to reduce the influence of partial occlusions in 3D face recognition, a novel 3D face recognition algorithm based on radial strings and local feature is proposed. Method This 3D face recognition algorithm includes four main parts. First, locating the nasal tip on 3D face data using shape index, extracting the radial strings after nasal tip is located, then uniform sampling on every radial strings. To fully use the neighbor information of radial strings, a radial string representation which encoded radial strings into local feature is proposed. In this algorithm, we extract three local features, including the center of every two adjacent sample points, the area of local region and the histogram of slant angle. Then local feature descriptors with these local features to represent local region are constructed. Second, sparse cloud points would lead to nonuniform sample points, it would lead to large errors in the matching result. To address this problem, an operator that merges adjacent local regions is adopted. Third, constructing cost function of the local feature on corresponding local region, and constructing similarity vectors of corresponding radial strings with this cost function. Fourth, matching corresponding radial strings according to these similarity vectors, and recognizing 3D face by the result of all radial strings. Result The experiments are conducted based on the FRGC v2.0 database and the Bosphorus database. FRGC v2.0 is a large-scale public 3D face database, including 466 subjects, 4007 3D point cloud totally. Bosphorus database is a new 3D face database, including 105 subjects,4666 3D point cloud totally, and this database including partial occlusions at different level. Because FRGC v2.0 database is standard and at high level, we choose 300 subjects with neutral and not occluded 3D face point cloud to test the recognition rate of different local feature, the rank-one recognition rate is 95.2%, is 0.9% and 2.4% higher than the other two local feature respectively. Although those local feature is only 0.9% higher than the second high local feature, those local feature promote the convenience of merge of adjacent local regions. Then we choose 300 3D face point cloud in Bosphorus database to construct the experiments of recognition rate and time with partial occlusions 3D face, the rank-one recognition rate is 92.0%, is 2.7%, 3.0% and 0.4% higher than the other three recognition methods respectively. As for time, it cost 8.17s, are the least in those recognition methods, and is 2.05s, 0.18s and 34.43s less than the other three methods respectively.In those experiments, our methods receive the best result both recognition rate and recognition time. Conclusion The proposed method of 3D face recognition based on radial strings and local feature extracts the adjacent information of radial strings effectively, then constructing cost function of corresponding local regions with the adjacent information to achieve region matching. The similarity vector that constructed by cost function of local feature reduces the influence of partial occlusions effectively. The result of experiments demonstrated that the algorithm which we proposed achieved a high recognition rates and robust to partial occlusions. This 3D face recognition method is suitable for 3D face recognition with partial occlusions. Because this method must locate the position of nasal tip, so it can not apply to the 3D face data which nasal tip is occluded.