基于支持向量机的颅骨性别识别-《计算机技术与发展》

文章信息/Info

Author(s):: YANG Wen; LIU Xiao－ning; ZHU Fei; School of Information Science and Technology，Northwest University，Xi’an 710127，China

Keywords:: sex determination; measurable features; non－measurable features; support vector machine; leave－out cross validation

摘要:: 颅骨性别识别在法医学和颅骨面貌复原等领域具有重要研究意义和应用价值。文中以新疆吐鲁番地区 117 例维吾尔族成人三维颅骨数字模型为研究对象，首先，对颅骨模型利用自主开发的系统标定 78 个特征点，其中 12 个位于颅骨正中矢状面、66 个对称分布于颅骨两侧;然后，提取可测量特征和非可测量特征，对可测量特征直接测量，对非可测量特征进行量化表示;最后，利用支持向量机方法对提取的特征向量进行降维并设计分类器，实现对颅骨的性别分类。实验结果表明，将测量特征和非可测量特征结合包含更多的性别识别信息，支持向量机的方法可以很好地实现性别分类，并能提高性别识别精度，利用留一交叉验证进行测试，其中男性识别正确率达 90．0%，女性识别正确率达 94．7%，平均识别正确率达92．4%。

Abstract:: Sex determination of skull has important research significance and application value in forensic medicine and skull rejuvenation．In this paper，we take the 117 cases of Uygur adult three－dimensional skull digital model in Turpan area of Xinjiang as the research ob-ject． First，78 features points of the skull model are calibrated by the independent development system，of which 12 are located in the me-dian sagittal plane of the skull and 66 are symmetrically distributed on both sides of the skull． Then，the measurable features and nonmeasurable features are extracted，and the former is measured directly and the latter is quantized． Finally，the support vector machinemethod is used to reduce the dimensions of the extracted eigenvectors and design the classifier to complete the classification of the skull． The experiment shows that combining the measured and non－measurable features with more sex determination information，SVM can a-chieve gender classification well and improve the accuracy of gender identification． Using leave－out cross validation test，the recognitionaccuracy of males is 90%，that of females is 94．7%，and that of the average 92．4%．