DFMFMA: Design of an efficient Deep learning model for age estimation from frontal Faces via Multimodal Feature representation & Multicomponent Augmented analysis

Abstract

Abstract Age estimation from facial images has various applications, including security, healthcare, and entertainment. Accurate age estimation is essential for age- dependent services such as age-restricted content filtering, targeted advertising, and personalized health care. However, age estimation from facial images is a challenging task due to various factors such as variations in pose, illumination, occlusion, and aging patterns. Conventional approaches for age estimation from facial images are typically based on handcrafted features, such as texture, shape, and appearance features. These approaches often suffer from limited discriminative power and robustness to variations in the images. With the advent of deep learning, there has been a surge of interest in using deep neural networks for age estimation from facial images. Deep neural networks can learn complex and discriminative features from the images, enhancing the accuracy and robustness of the age estimation models. The proposed approach in this paper utilizes a deep learning-based approach for age estimation from frontal face images. The approach involves the analysis of facial components, including eyes, nose, and mouth, to capture age-related changes in different regions of the face images. The components are augmented using various operations such as rotation and shifting to improve the robustness of the model against variations in pose, illumination, and occlusions. The augmented components are then converted into multimodal features and individually classified using an efficient & novel Binary Cascaded CNN that employs binary weights and activations, reducing the model’s complexity and improving its efficiency levels. The use of multimodal features allows the model to capture the age-related changes in multiple domains, improving the dis- criminative efficiency of the model under multiple class scenarios. The + model’s accuracy is evaluated on augmented FGNET datasets and samples, achieving an accuracy of 99.5% with an MAE of 1.26 across all age groups. The high accuracy achieved by the proposed model highlights its effectiveness and potential for real-world age estimation scenarios.