Fingerprint Based Male-Female Classification Manish Verma and Suneeta Agarwal Computer Science Department, Motilal Nehru National Institute of Technology Allahabad Uttar Pradesh India manishverma649@gmail.com, suneeta@mnnit.ac.in Abstract. Male-female classification from a fingerprint is an important step in forensic science, anthropological and medical studies to reduce the efforts required for searching a person. The aim of this research is to establish a relationship between gender and the fingerprint using some special features such as ridge density, ridge thickness to valley thickness ratio (RTVTR) and ridge width. Ahmed Badawi et. al. showed that male-female classification can be done correctly upto 88.5% based on white lines count, RTVTR & ridge count using Neural Network as Classifier. We have used RTVTR, ridge width and ridge density for classification and SVM as classifier. We have found male-female can be correctly classified upto 91%. Keywords: gender classification, fingerprint, ridge density, ridge width, RTVTR, forensic, anthropology. 1 Introduction For over centuries, fingerprint has been used for both identification and verification because of its uniqueness. A fingerprint contains three level of information. Level 1 features contain macro details of fingerprint such as ridge flow and pattern type e.g. arch, loop, whorl etc. Level 2 features refer to the Galton characteristics or minutiae, such as ridge bifurcation or ridge termination e.g. eye, hook, bifurcation, ending etc. Level 3 features include all dimensional attributes of ridge e.g. ridge path deviation, width, shape, pores, edge contour, ridges breaks, creases, scars and other permanent details [10]. Till now little work has been done in the field of male-female fingerprint classification. In 1943, Harold Cummnins and Charles Midlo in the book “Fingerprints, Palm and Soles” first gave the relation between gender and the fingerprint. In 1968, Sarah B Holt, Charles C. Thomas in the book “The Genetics of the Dermal Ridges” gave same theory with little modification. Both state the same fact that female ridges are finer/smaller and have higher ridge density than males. Acree showed that females have higher ridge density [9]. Kralik showed that males have higher ridge width [6]. Moore also carried out a study on ridge to ridge distance and found that mean distance is more in male compared to female [7]. Dr. Sudesh Gungadin showed that a ridge count of ≤13 ridges/25 mm2 is more likely to be of males and that of ≥14 ridges/25 mm2 is likely to be of females [2]. Ahmed Badawi et. al. showed that male-female can be correctly classified upto 88.5% [1] based on white lines count, RTVTR & ridge count using Neural Network as Classifier. According to the research of E. Corchado et al. (Eds.): CISIS 2008, ASC 53, pp. 251–257, 2009. © Springer-Verlag Berlin Heidelberg 2009 springerlink.com 252 M. Verma and S. Agarwal Cummnins and Midlo, A typical young male has, on an average, 20.7 ridges per centimeter while a young female has 23.4 ridges per centimeter [8]. On the basis of studies made in [6], [1], [2], ridge width, RTVTR and ridge density are significant features for male-female classification. In this paper, we studied the significance of ridge width, ridge density and ridge thickness to valley thickness ratio (RTVTR) for the classification purpose. For classification we have used SVM classifier because of its significant advantage. Artificial Neural Networks (ANNs) can suffer from multiple local minima, the solution with an SVM is global and unique. Unlike ANNs, the computational complexity of SVMs does not depend on the dimensionality of the input space. ANNs use empirical risk minimization, whilst SVMs use structural risk minimization. SVMs are less prone to overfitting [13]. 2 Materials and Methods In our Male-Female classification analysis with respect to fingerprints, we extracted three features from each fingerprint. The features are ridge width, ridge density and RTVTR. Male & Female are classified using these features with the help of SVM classifier. 2.1 Dataset We have taken 400 fingerprints (200 Male & 200 Female) of indian origin in the age group of 18-60 years. These fingerprint are divided into two disjoint set for training and testing, each set contains 100 male and 100 female fingerprints. 2.2 Fingerprint Feature Extraction Algorithm The flowchart of the Fingerprint Feature Extraction and Classification Algorithm is shown in Fig. 1. The main steps of the algorithm are: Normalization [4] Normalization is used to standardize the intensity values of an image by adjusting the range of its grey-level values so that they lie within a desired range of values e.g. zero mean and unit standard deviation. Let I(i,j) denotes the gray-level value at pixel (i,j), M & VAR denote the estimated mean & variance of I(i,j) respectively & N(i,j) denotes the normalized gray-level value at pixel (i,j). The normalized values is defined as follows: (1) Where M0 and VAR0 are desired mean and variance values respectively. Image Orientation [3] Orientation of a fingerprint is estimated by the least mean square orientation estimation algorithm given by Hong et. al. Given a normalized image, N, the main steps of Fingerprint Based Male-Female Classification 253 Fig. 1. Flow chart for Fingerprint Feature Extraction and Classification Algorithm the orientation estimation are as follows: Firstly, a block of size wXw (25X25) is centred at pixel (i, j) in the normalized fingerprint image. For each pixel in this block, compute the Gaussian gradients ∂x(i, j) and ∂y(i, j), which are the gradient magnitudes in the x & y directions respectively. The local orientation of each block centered at pixel (i, j) is estimated using the following equations [11]. (2) (3) (4) 254 M. Verma and S. Agarwal where θ(i,j) is the least square estimate of the local orientation at the block centered at pixel (i,j). Now orient the block with θ degree around the center of the block, so that the ridges of this block are in vertical direction. Fingerprint Feature Extraction In the oriented image, ridges are in vertical direction. Projection of the ridges and valleys on the horizontal line forms an almost sinusoidal shape wave with the local minima points corresponding to ridges and maxima points corresponds to valleys of the fingerprint. Ridge Width R is defined as thickness of a ridge. It is computed by counting the number of pixels between consecutive maxima points of projected image, number of 0’s between two clusters of 1’s will give ridge width e.g. 11110000001111 in above example, ridge width is 6 pixels. Valley Width V is defined as thickness of valleys. It is computed by counting the number of pixels between consecutive minima points of projected image, number of 1’s between two clusters of 0’s will give valley width e.g. 00001111111000 in above example, valley width is 7 pixels. Ridge Density is defined as number of ridges in a given block. e.g. 001111100011111011 Above string contains 3 ridges in a block. So ridge density is 3. Ridge Thickness to Valley Thickness Ratio (RTVTR) is defined as the ratio of ridge width to the valley width and is given by RTVTR = R/V. Fig. 2. Segmented image is Oriented and then projected to line from its binary transform we got ridge and valley width Example 1. Fig. 2. shows a segment of normalized fingerprint, which is oriented so that the ridges are in vertical direction. Then these ridges are projected on horizontal line. In the projected image, black dots show a ridge and white show a valley. Fingerprint Based Male-Female Classification 255 Classification SVM’s are used for classification and regression. SVM’s are set of related supervised learning methods. A special property of SVMs is that they simultaneously minimize the empirical classification error and maximize the geometric margin; hence they are also known as maximum margin classifiers. For a given training set of instance-label pairs (Xi, yi), i=1..N, where N is any integer showing number of training sample, Xi ∈Rn (n denotes the dimension of input space) belongs to the two separating classes labeled by yi∈{-1,1}, This classification problem is to find an optimal hyperplane WTZ + b =0 in a high dimension feature space Z by constructing a map Z=φ(X) between Rn and Z. SVM determines this hyperplane by finding W and b which satisfy (5) . Where ξ i ≥ 0 and yi[W φ(Xi)+b] ≥ 1- ξ i holds. Coefficient c is the given upper bound and N is the number of samples. The optimal W and b can be found by solving the dual problem of Eqn. (5), namely T (6) . Where 0 ≤ αi ≤ c (i = 1,….,N) is Lagrange multiplier and it satisfies 0. ) and we adopt the RBF function to map the input vecLet tors into the high dimensional space Z. The RBF Function is given by . (7) Where γ= 0.3125,c=512.Values of c & γ are computed by grid search [5]. The decision function of the SVM classifier is presented as (8) Where K(.,.) is the kernel function, which defines an inner product in higher dimension space Z and it satisfies that . The decision function sgn(φ) is the sign function and if φ≥0 then sgn(φ)=1 otherwise sgn(φ)=-1 [12]. 3 Results Our experimental result showed that if we consider any single feature for Male– Female classification then the classification rate is very low. Confusion matrix for Ridge Density (Table 1), RTVTR (Table 2) and Ridge Width (Table 3) show that their classification rate is 53, 59.5 and 68 respectively for testing set. But by taking all these features together we obtained the classification rate 91%. Five fold cross validation is used for the evaluation of the model. For testing set, Combining all these features together classification rate is 88% (Table 4). 256 M. Verma and S. Agarwal Table 1. Confusion Matrix for Male-Female classification based on Ridge Density only for Testing set Actual\Estimated Male Female Total Male 47 41 88 Female 53 59 112 Total 100 100 200 For Ridge Density the classification rate is 53% Table 2. Confusion Matrix for Male-Female classification based on RTVTR only for Testing set Actual\Estimated Male Female Total Male 30 11 41 Female 70 89 159 Total 100 100 200 For RTVTR the classification rate is 59.5% Table 3. Confusion Matrix for Male-Female classification based on Ridge Width only for Testing set Actual\Estimated Male Female Total Male 51 15 66 Female 49 85 134 Total 100 100 200 For Ridge Width the classification rate is 68% Table 4. Confusion Matrix for Male-Female classification based on combining Ridge Density, Ridge Width and RTVTR only for Testing set Actual\Estimated Male Female Total Male 86 10 96 Female 14 90 104 Total 100 100 200 For Testing set the classification rate is 88% 4 Conclusion Accuracy of our model obtained by five fold cross validation method is 91%. Our results have shown that the ridge density, RTVTR and Ridge Width gave gave 53%, 59.5% and 68% classification rates respectively. Combining all these features together, we obtained 91% classification rate. Hence, our method gave 2.5 % better result than the method given by Ahmed Badawi et al. Fingerprint Based Male-Female Classification 257 References 1. Badawi, A., Mahfouz, M., Tadross, R., Jantz, R.: Fingerprint Based Gender Classification. In: IPCV 2006, June 29 (2006) 2. Sudesh, G.: Sex Determination from Fingerprint Ridge Density. Internet Journal of Medical Update 2(2) (July-December 2007) 3. Hong, L., Wan, Y., Jain, A.K.: Fingerprint Image Enhancement: Algorithms and Performance Evaluation. IEEE Trans. Pattern Analysis and Machine Intelligence 20(8), 777–789 (1998) 4. Raymond thai, Fingerprint Image Enhancement and Minutiae Extraction (2003) 5. Hsu, C.-W., Chang, C.-C., Lin, C.-J.: A practical guide to support vector classification, http://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf 6. Kralik, M., Novotny, V.: Epidermal ridge breadth: an indicator of age and sex in paleodermatoglyphics. Variability and Evolution 11, 5–30 (2003) 7. Moore, R.T.: Automatic fingerprint identification systems. In: Lee, H.C., Gaensslen, R.E. (eds.) Advances in Fingerprint Technology, p. 169. CRC Press, Boca Raton (1994) 8. Cummins, H., Midlo, C.: Fingerprints, Palms and Soles. An introduction to dermatoglyphics, p. 272. Dover Publ., New York (1961) 9. Acree M.A.: Is there a gender difference in fingerprint ridge density? Federal Bureau of Investigation, Washington, DC 20535-0001, USA 10. Jain, A.K., Chen, Y., Demerkus, M.: Pores and Ridges: High-Resolution Fingerprint Matching Using Level 3 Fetures. IEEE Transaction on Pattern Analysis and Matching Intelligence 29 (January 2007) 11. Rao, A.: A Taxonomy for Texture Description and Identification. Springer, New York (1990) 12. Ji, L., Yi, Z.: SVM-based Fingerprint Classification Using Orientation Field. In: Third International Conference on Natural Computation (ICNC 2007) (2007) 13. Support Vector Machines vs Artificial Neural Networks, http://www.svms.org/anns.html
© Copyright 2025 Paperzz