Facial Image Pre-processing for Enhanced Ethnicity Classification

Facial recognition and ethnicity detection is an important task in various applications such as security, marketing, and social sciences. However, the accuracy of Face recognition and ethnicity detection algorithms can be impacted by various factors such as lighting conditions, noise and irrelevant features in image background, variation in facial expressions, face size, pose and orientation across images. In this study, we present a novel face image preprocessing model in this work that aims to enhance the performance of deep learning models like ResNet, DenseNet and other state of the art Deep Learning models on facial recognition tasks, specifically ethnicity detection by addressing these issues. Our novel method for preprocessing facial images incorporates several techniques to enhance the quality of the input images. Firstly, we removed the background using DeepLab-v3. Secondly, we employed facial landmarks detection and facial pose and orientation correction using facial alignment based on the angle between the eyes. Thirdly, to facilitate race classification, we performed facial cropping to isolate the facial region and other critical features essential for the task. Finally, we ensured consistent lighting conditions across all images using the Multi-scale retinex algorithm. These techniques together enabled the deep learning model to better analyze and classify facial data. To assess the efficacy of our model, we carried out a series of experiments on the UTK Face dataset and compared the outcomes with those obtained when our preprocessing model was not utilized.The experimental outcomes show that our suggested pre-processing model improves the f1-score of ethnicity detection classifiers by up to 6%. The proposed facial image preprocessing model is efficient, fast, and easy to implement, enabling real-time applications, making it appropriate. Our research highlights the importance of preprocessing techniques in enhancing ethnicity detection performance and provides a practical solution for improving the accuracy of ethnicity detection and facial recognition systems.