PREDICTION OF GEOMETRIC FEATURES IN 3D FACIAL SOFT TISSUE IMAGES USING FULLY CONNECTED NEURAL NETWORK

The relevant geometric prominence of the three-dimensional facial soft tissue images is a crucial factor in forensic reconstruction, craniofacial surgery, and biometric identification. The image processing technique is one of the most critical steps in analysing 3D facial images. Traditional image processing methods often face problems while capturing the soft tissue image's non-linear deformations and complex anatomical variation. So, this paper presented a solution to this challenge by formulating an accurate geometric feature prediction model using Fully Connected Neural Network (FCNN) for 3D soft tissue facial images. The important feature of the model is that it was trained on high-definition 3D face scans and leveraged deep learning-driven feature extraction to increase the model's accuracy. Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Structural Similarity Index (SSIM) are the metrics used for evaluation. The presented model performed better than traditional deep learning models based on the results. Finally, the proposed model has provided far more accurate results in shape prediction, which makes this model highly suitable for application in personalized facial reconstruction, clinical diagnostics, and aesthetic surgery planning. These enhancements in the technique contribute to the development of an improved automated 3D facial analysis, making a better way of modelling the soft tissue area and the computational facial anatomy.