Face Recognition and Detection using CNN and Cascade Detection

Face biometrics has been used extensively in various spheres of the current technological world due to its high accuracy and non-intrusive nature. In this study, we present a comprehensive system built, leveraging the power of Cascade Detector and Convolutional Neural Networks (CNNs) for accurate and efficient face detection and recognition, with a specific focus on individuals of Asian ethnicity. The face detection component employs a Cascade Detector to identify key facial features, including the eyes, nose, and mouth. Simultaneously, we utilize three prominent CNN models, namely AlexNet, SqueezeNet, and GoogleNet, for recognizing faces in RGB images. Additionally, a custom-built CNN model is employed for grayscale images, enhancing the versatility of the system across different imaging scenarios. The models are trained on a diverse dataset that includes individuals of Asian ethnicity, enabling the system to excel in recognizing facial features unique to this demographic. Evaluation metrics including Accuracy, False Acceptance Rate (FAR) and False Rejection Rate (FRR) were computed across diverse datasets, facilitating a comprehensive analysis of model performance. Among the colour datasets, the Indian female dataset consistently yielded the highest accuracy, ranging from 65% to 95%, while the Southeast Asian dataset exhibited comparatively lower accuracy, ranging from 40% to 60%. The Indian male dataset fell within the range of 55% to 75% accuracy, grayscale datasets, our custom-built CNN model achieved impressive accuracies of 92.68%.