CRNet: Convolutive Recurrent Network for Suspect Face Identification

Identifying suspects in critical situations-particularly when they are wearing scarves, masks, or are in environments with light obstructions and concealed facial expressions-poses significant challenges. To address these issues, a method known as the Convolutive Recurrent Network (CRNet) for suspect face identification is proposed. CRNet utilises deep neural networks, specifically the Residual Network-50, leveraging a transfer learning approach for efficient feature extraction. In addition, Bidirectional Long Short-Term Memory (BiLSTM) layers are employed to capture spatial and recurrent features, with BiLSTM layers serving as the core component of the model. CRNet is designed to overcome the limitations of current models in managing complex situations, such as scarves, spectacles, high illumination, and varied expressions. CRNet fills this gap by integrating mechanisms that provide flexibility for ambiguous features and variable lighting conditions. Experimental and comparative analysis demonstrates that CRNet significantly outperforms existing methods, providing notable improvements in both accuracy and reliability. This approach introduces a rapid feature-learning method for precise suspect identification by integrating spatial dependencies, enhancing versatility across various computer vision domains. The model’s potential impact on criminal investigations is substantial due to its fast bidirectional feature processing. Experimental results demonstrate the robustness and adaptability of CRNet, achieving accuracy rates of 97.46% on the Extended Cohn-Kanade dataset, 98.08% on the Augmented Reality dataset, and 99.58% on the Extended Yale B dataset-substantially surpassing the baseline accuracy of 46.00%.

N/A

ANTIC 2024: 4th International Conference on Advanced Network Technologies and Intelligent Computing

2334

9783031837890

N/A

N/A

15