Unconstrained Gender Recognition from Periocular Region Using Multiscale Deep Features

The gender recognition problem has attracted the attention of computer vision community due to its importance in many applications (e.g., surveillance and human–computer interaction [HCI]). Images varying levels illumination, occlusion, other factors are captured uncontrolled environments. Iris facial technology cannot be used on these images because iris texture is unclear instances, faces may covered by a scarf, hijab, or mask COVID-19 pandemic. periocular region reliable source information it features rich discriminative biometric features. However, most existing classification approaches have been designed based hand-engineered validated controlled Motivated superior performance deep learning, we proposed new method, PeriGender, inspired design principles ResNet DenseNet models, that can classify using from region. system utilizes dense concept residual model. Through skip connections, reuses different scales strengthen Evaluations challenging datasets indicated outperformed state-of-the-art methods. It achieved 87.37%, 94.90%, 94.14%, 99.14%, 95.17% accuracy GROUPS, UFPR-Periocular, Ethnic-Ocular, IMP, UBIPr datasets, respectively, open-world (OW) protocol. further 97.57% 93.20% for adult GROUPS dataset closed-world (CW) OW protocols, respectively. results showed middle between eyes plays crucial role masculine features, feminine identified through eyebrow, upper eyelids, corners eyes. Furthermore, whole without cropping enhances PeriGender’s learning capability, improving understanding both eyes’ global structure discontinuity.