Integrative Deep-Ridge Architecture for Intelligent Fingerprint-based Gender Recognition

Fingerprint-based gender identification has emerged as an important area of investigation in biometric studies owing to its relevance to various applications, including forensic analysis, biometric database management, and identity verification. This paper proposes a hybrid framework for fingerprint-based gender classification, which combines the effectiveness of handcrafted feature extraction methods with those of deep learning-based representations. Fingerprint images are classified into male and female classes, and several feature extraction methods are used to extract features from fingerprint images. Ridge-based statistical features are used, which are obtained from grayscale finger-print images. Ridge density, contrast, energy, and area features are used because they vary between male and female classes. Moreover, texture features are also used, which are obtained by applying Gabor filters at various orientations and frequencies. For improving the discriminative power of the features, a convolutional neural network (CNN) is used. Specifically, a pre-trained ResNet50 architecture is used as a feature extraction tool, and the fully connected layers are removed. Global average pooling is used to extract compact feature vectors. The extracted ridge features, texture descriptors, and deep learning features are normalized and fused at the feature level to create a comprehensive feature representation of fingerprint information. The feature vectors are classified using a Support Vector Ma-chine classifier with a radial basis function (RBF) kernel. The experimental evaluation of the proposed method shows that the method successfully utilizes the fingerprint information at the local structural level as well as the feature level, thus enhancing the gender classification performance.