基于拇指与食指指纹组合特征分析人的性别及年龄

doi:10.16359/j.1000-3193/AAS.2023.0043

摘要/Abstract

摘要：

利用指纹特征分析人的性别及年龄，一直是法医学和人类学领域的挑战性项目。现有研究因统计样本量少、特征利用率低、模型学习能力低等不足，导致利用指纹特征分析性别及年龄的准确率低。本文从多种分类及回归机器学习模型比对分析的角度，对2980份拇指及食指指纹捺印样本（男性1500份，女性1480份）的指纹特征进行统计测量，并在拇指及食指不同指纹特征相互组合的情况下，观察各机器学习模型的性别分类、年龄回归的准确率。结果表明，同时使用拇指及食指指纹特征进行学习的性别分类、年龄回归的准确率高于使用单一手指指纹特征的准确率。其中，使用指纹特征对性别进行分类的结果中，F₁衡量指标值最高为0.979；对男性指纹样本进行年龄回归的结果中最高准确率为86.7%，女性指纹样本年龄回归的最高准确率为85.3%，证明使用拇指及食指指纹特征综合进行学习，可提高性别分类及年龄回归准确率思路的有效性。

关键词: 指纹, 机器学习, 性别, 年龄

Abstract:

The analysis of gender and age by using fingerprint characteristics has been a persistent challenge in the fields of forensic medicine and anthropology. Relevant studies have many shortcomings and deficiencies that need to be improved. For example, the number of fingerprint samples is insufficient, the utilization rate of fingerprint characteristics is low, and the learning ability of the model used is low, all of which will lead to the decline of the accuracy of using fingerprint characteristics to analyze individual gender and age. To address these limitations, this study adopts a multi-classification approach and performs a comparative analysis of regression machine learning models. A comprehensive analysis was conducted on a dataset comprising 2980 thumb and forefinger fingerprint samples, consisting of 1500 male and 1480 female individuals. Statistical measurements were performed on the fingerprint characteristics of the dataset, and the accuracy of gender classification and age regression for each machine learning model was observed. The models were tested by combining different fingerprint characteristics from the thumb and forefinger. The findings reveal that utilizing both thumb and forefinger fingerprint characteristics significantly improves the accuracy of gender classification and age regression compared to using single finger fingerprint characteristics. Notably, the highest F₁ score achieved for gender classification using fingerprint characteristics was 0.979, indicating a remarkably high accuracy level. For male fingerprint samples, the highest accuracy in age regression reached 86.7%, while for female fingerprint samples, it yielded a highest accuracy of 85.3%. These outcomes validate the efficacy of comprehensive learning with thumb and forefinger fingerprint characteristics in enhancing gender classification and age regression accuracy. The results of this study contribute significant insights to the application of fingerprint characteristics in determining gender and age. By addressing the limitations of previous research and emphasizing the importance of multi-classification and comparative analysis, it demonstrates the potential for achieving higher accuracy in gender and age analysis through the integration of thumb and forefinger fingerprint characteristics. These findings hold profound implications for the fields of forensic medicine and anthropology, offering valuable support for future research and practical applications in fingerprint feature analysis. With continued advancements in technology and further research, it is anticipated that the application of fingerprint characteristics in gender and age analysis will continue to evolve and improve. The comprehensive understanding derived from this study serves as a foundation for future investigations, encouraging the exploration of enhanced methodologies and refining the accuracy and reliability of gender and age analysis through fingerprint characteristics.

Key words: fingerprint, machine learning, gender, age

中图分类号:

Q983

赵瑞敏, 刘凯, 孙鹏, 张忠良. 基于拇指与食指指纹组合特征分析人的性别及年龄[J]. 人类学学报, 2024, 43(03): 427-439.

ZHAO Ruimin, LIU Kai, SUN Peng, ZHANG Zhongliang. Using fingerprint characteristics of a person’s thumb and forefinger to analyze his gender and age[J]. Acta Anthropologica Sinica, 2024, 43(03): 427-439.

图/表 11

图1 指纹特征测量示例

Fig.1 Fingerprint feature measurement example diagram

图2 同一男性不同年龄阶段的指纹变化 A-F表示不同男性的指纹 A-F indicates different male fingerprints；下标1、2表示拇指指纹 subscript 1, 2 indicates thumb fingerprints；下标3、4表示食指指纹 subscript 3, 4 indicates forefinger fingerprints

Fig.2 Comparison of fingerprint changes in the same person at different ages

图3 男性和女性的拇指和食指指纹特征全年龄段测量数据分布

Fig.3 Distribution of thumb and forefinger fingerprint features of males and females in all-age

图4 男性和女性的拇指和食指指纹特征随年龄变化规律图 (a)-(i). 分别对应男性和女性的拇指和食指的dr、br、bv、nh、nv、L、ni、nf和P随年龄变化的模式，图中的每个特征值都是在每个年龄段测量的数据的中值，MT、FT、MF、FF 分别表示男性拇指、女性拇指、男性食指和女性食指

Fig.4 Patterns of thumb and forefinger fingerprint features with age in males and females correspond to the pattern of changes in dr, br, bv, nh, nv, L, ni, nf and P with age in the thumb and forefinger for males and females,and each feature value in the figure shows the median value of the data measured at each age, MT, FT, MF, FF stand for male thumb, female thumb, male index finger, and female index finger

表1 不同特征组合在10种分类方法下的F1分数

Tab.1 F1 scores of different feature combinations based on 10 classification methods

方法Method→ 分组groups↓	RAF	ADB	CTB	EXT	KNN	BPN	SVM	XGB	LGB	BYS
1	0.949	0.951	0.954	0.941	0.938	0.907	0.550	0.942	0.948	0.910
2	0.951	0.954	0.956	0.950	0.946	0.909	0.583	0.949	0.955	0.901
3	0.956	0.955	0.955	0.949	0.944	0.911	0.648	0.950	0.961	0.915
4	0.958	0.954	0.957	0.956	0.949	0.924	0.642	0.957	0.959	0.912
5	0.958	0.954	0.962	0.955	0.953	0.923	0.557	0.953	0.964	0.921
6	0.963	0.960	0.962	0.957	0.951	0.928	0.751	0.961	0.964	0.914
7	0.959	0.963	0.966	0.964	0.957	0.929	0.547	0.961	0.966	0.930
8	0.964	0.959	0.971	0.964	0.953	0.925	0.710	0.962	0.967	0.916
9	0.975	0.967	0.971	0.963	0.954	0.929	0.706	0.957	0.973	0.924

表2 3类特征在10种分类方法下的F1分数

Tab.2 F1 scores of 3 types of features under 10 classification methods

特征Feature→ 方法 Method↓	拇指Thumb	食指Forefinger	拇指食指组合特征Thumb & Forefinger
RAF	0.975	0.966	0.979
ADB	0.967	0.963	0.969
CTB	0.971	0.963	0.973
EXT	0.963	0.951	0.967
KNN	0.954	0.938	0.962
BPN	0.939	0.921	0.941
SVM	0.716	0.679	0.712
XGB	0.957	0.934	0.965
LGB	0.973	0.956	0.954
BYS	0.914	0.905	0.938

图5 性别分类可视化分析 A.男性指纹 Male fingerprints; B.被误分类至女性指纹的男性指纹 Male fingerprints that were misclassified as female fingerprints; C. 女性指纹 Female fingerprints

Fig.5 Visual analysis of gender classification

图6 RAF方法参数调优图 A-F. RAF方法在A到F的6个特征组合中随着参数的变化而产生的回归精度

Fig.6 RAF method parameter tuning diagram the regression accuracy of the RAF method with the variation of parameters for 6 combinations of features from A to F

图7 8种回归方法在不同特征组合下的回归准确率

Fig.7 Age regression using each of the eight methods based on parameter settings under different combinations of features

表3 在有无性别分类的情况下分别使用RAF得到的年龄回归准确率

Tab. 3 Accuracy of RAF age regression without gender classification and after gender classification

特征Feature→	男性Male			女性Female
方法Method↓	拇指Thumb	食指Forefinger	拇指食指组合特征Thumb& Forefinger	拇指Thumb	食指Forefinger	拇指食指组合特征Thumb& Forefinger
RAF	75.8%	72.1%	82.9%	78.5%	73.6%	84.5%
ADB	75.2%	72.3%	86.7%	77.1%	74.5%	85.3%
EXT	74.6%	71.1%	85.3%	77.8%	72.4%	81.3%
CTB	71.3%	69.2%	81.6%	75.7%	70.9%	81.1%
KNN	67.8%	65.5%	72.1%	73.2%	68.3%	78.9%
SVR	59.6%	53.7%	68.4%	73.4%	70.6%	80.8%
LGB	73.2%	72.0%	84.2%	75.6%	71.4%	81.2%
MLR	68.5%	67.3%	75.2%	71.3%	69.3%	80.5%

图8 年龄回归可视化分析

Fig.8 Age regression visual analysis

参考文献 29

[1]	Champod C, Lennard CJ, Margot P, et al. Fingerprints and other ridge skin impressions[M]. La Paz: CRC Press, 2004
[2]	Monson KL, Roberts MA, Knorr KB, et al. The permanence of friction ridge skin and persistence of friction ridge skin and impressions: A comprehensive review and new results[J]. Forensic science international, 2019, 297: 111-131 doi: S0379-0738(19)30039-8 pmid: 30784948
[3]	Gorka M, Thomas A, Bécue A. Chemical composition of the fingermark residue: Assessment of the intravariability over one year using MALDI-MSI[J]. Forensic Science International, 2022, 338: 111380
[4]	Mundorff AZ, Bartelink EJ, Murad TA. Sexual dimorphism in finger ridge breadth measurements: a tool for sex estimation from fingerprints[J]. Journal of forensic sciences, 2014, 59(4): 891-897 doi: 10.1111/1556-4029.12449 pmid: 24673752
[5]	Nayak VC, Rastogi P, Kanchan T, et al. Sex differences from fingerprint ridge density in Chinese and Malaysian population[J]. Forensic science international, 2010, 197(1-3): 67-69 doi: 10.1016/j.forsciint.2009.12.055 pmid: 20071110
[6]	Sánchez-andrés A, Barea JA, Rivaldería N, et al. Impact of aging on fingerprint ridge density: Anthropometry and forensic implications in sex inference[J]. Science & justice, 2018, 58(5): 323-334
[7]	Kanchan T, Krishan K. Anthropometry of hand in sex determination of dismembered remains-A review of literature[J]. Journal of Forensic and Legal Medicine, 2011, 18(1): 14-17 doi: 10.1016/j.jflm.2010.11.013 pmid: 21216373
[8]	Ceyhan EB, Saĝiroĝlu Ş, Tatoĝlu S, et al. Age estimation from fingerprints. Examination of the population in Turkey[A]. In: Wani M(Ed.). proceedings of the 2014 13^th International Conference on Machine Learning and Applications[C]. Detroit: IEEE, 2014, 478-481
[9]	宋焕庭, 唐玮, 张丽梅, 等. 指纹与年龄相关性的量化分析[J]. 人类学学报, 2022, 41(6): 1047-1057
[10]	Falohun A, Fenwa O, Ajala F. A fingerprint-based age and gender detector system using fingerprint pattern analysis[J]. International Journal of Computer Applications, 2016, 136(4): 0975-8887
[11]	Jayakala G, Sudha L. Fingerprint analysis for age estimation using deep learning models (ResNet50 and VGG-16)[J]. International Journal of Health Sciences, 2022, 6(S3): 6781-6789
[12]	Rivaldería N, Gutiérrez-Redomero E. Interpopulational differences in the frequency and distribution of delta types[J]. Forensic Science International, 2021, 322: 110749
[13]	Dao JRH, Kazin RA. Gender differences in skin: a review of the literature[J]. Gender medicine, 2007, 4(4): 308-328 doi: 10.1016/s1550-8579(07)80061-1 pmid: 18215723
[14]	Farage MA, Miller KW, Elsner P, et al. Characteristics of the aging skin[J]. Advances in wound care, 2013, 2(1): 5-10 doi: 10.1089/wound.2011.0356 pmid: 24527317
[15]	Da Silva Carvalho D, De Alecrim MMV, De Sousa Júnior R T, et al. Outcome of sex determination from ulnar and radial ridge densities of Brazilians' fingerprints: applying an existing method to a new population[J]. Science & Justice, 2022, 62(2): 181-192
[16]	Goutte C, Gaussier E. A Probabilistic Interpretation of Precision, Recall and F-Score, with Implication for Evaluation[A]. In: Losada (Ed.). Advances in Information Retrieval. ECIR 2005. Lecture Notes in Computer Science[C]. Heidelberg: Springer, 2005:345-359
[17]	Patil A, Singh S. Prakash C(Ed.). Differential private random forest[A]. In: Prakash C(Ed.). proceedings of the 2014 International Conference on Advances in Computing, Communications and Informatics (ICACCI)[C]. Delhi: IEEE, 2014, 2623-2630
[18]	Schapire RE. Explaining adaboost[J]. Empirical Inference: Festschrift in Honor of Vladimir N Vapnik, 2013, 37-52
[19]	Prokhorenkova L, Gusev G, Vorobev A, et al. CatBoost: unbiased boosting with categorical features[J]. Advances in neural information processing systems, 2018, doi:10.48550/arXiv.1706.09516
[20]	Geurts P, Ernst D, Wehenkel L. Extremely randomized trees[J]. Machine learning, 2006, 63: 3-42
[21]	Peterson LE. K-nearest neighbor[J]. Scholarpedia, 2009, 4(2): 1883
[22]	Buscema M. Back propagation neural networks[J]. Substance use & misuse, 1998, 33(2): 233-270
[23]	Hearst MA, Dumais ST, Osuna E, et al. Support vector machines[J]. IEEE Intelligent Systems and their applications, 1998, 13(4): 18-28
[24]	Chen T, Guestrin C. Xgboost: A scalable tree boosting system[J]. ACM, 2016. doi:10.1145/2939672.2939785
[25]	Ke G, Meng Q, Finley T, et al. Lightgbm: A highly efficient gradient boosting decision tree[A]. In: Von Luxburg U(Ed.). Advances in neural information processing systems 30 (niP2017)[C]. Long Beach: Neural Information Processing Systems Foundation, 2017
[26]	Rish I. An empirical study of the naive Bayes classifier[J]. Journal of Universal Computer Science, 2001, 1(2): 127
[27]	Smola AJ, Schölkopf B. A tutorial on support vector regression[J]. Statistics and computing, 2004, 14: 199-222
[28]	Eberly LE. Multiple linear regression[J]. Topics in Biostatistics, 2007, 165-187
[29]	陆宏, 霍正浩, 钟慧军, 等. 宁夏回、汉族指纹白线的研究[J]. 人类学学报, 2005, 24(1): 77-81