Exploring FMU-NET: A Comprehensive Study on Person Identification Through Novel Semantic Segmentation Techniques

The paper proposes novel face semantic segmentation for person identification. Various face segmentation procedures are used to analyze the accuracy of the model to learn specific patterns from the annotated masks. The FMU- Net, a simplistic semantic segmentation model, amalgamates components from both the Residual U-Net and Mobile U-Net architectures. It adeptly capitalizes on the strengths of these frameworks while honing in on extracting precise facial features from newly devised masks. This meticulous feature extraction strategy empowers the FMU-Net to surpass its counterparts, as evidenced by the comprehensive comparative analysis presented in the paper. In the initial phase, semantic maps serve as the foundation for training the model, and subsequently, test semantic maps are generated and juxtaposed across various neural network architectures. The second stage of the methodology encompasses the integration of a classification layer with the semantic segmentation model. This combined approach not only delineates semantic segments but also discerns the individual’s identity, achieved through rigorous evaluation metrics such as the F1 score. Central to this research is the quest for the optimal neural network architecture tailored for semantic segmentation in the realm of face recognition. The study delves into intricate nuances of semantic face recognition, wherein the extraction of nuanced facial features plays a pivotal role. By strategically employing semantic maps and leveraging the fusion of proven architectural elements, the FMU-Net stands as a testament to the advancements in this domain, showcasing its prowess in discerning intricate facial details for robust and accurate face recognition applications. Additionally, the paper aims to enhance the chosen segmentation model by improving its time efficiency and reducing its complexity. This is achieved through the utilization of novel annotated semantic masks, which provide valuable insights for modifying and optimizing the segmentation model.