Optimizing dense face alignment in Variations in illumination

Face alignment is an important problem in computer vision and has many uses in facial identification, facial expression analysis, and 3D face modeling. However, the performance of dense face alignment is often affected by changes in illumination and lighting conditions. In this paper, we propose an improved version of the 3DDFA-V3 model to improve the performance of the model under extreme illumination conditions. This approach combines improved data preprocessing, exerting adaptive learning rate to progress the learning process, and a greater number of datasets. All these improvements combined enhance the model’s robustness to variations in illumination. The experimental results show that the proposed method improves the performance by 26% in terms of Mean Squared Error (MSE), 21% in terms of Normalized Mean Error (NME), and 18% in terms of Structural Similarity Index Measure (SSIM). According to the results indicated, the 3DDFA-V3 model with optimization is better than the base model, and it is superior in terms of stability, consistency as well as its ability to provide accurate alignment in different lighting conditions.