Treffer: Criminal Face Detection using Deep Learning Algorithms.

Criminal face detection is an essential task in surveillance and security systems, aiming to accurately identify individuals from facial images. However, this process faces numerous challenges, including variations in facial expressions, lighting conditions, and occlusions, which can reduce the effectiveness of traditional deep learning models. This research presents ScanNet, a convolutional neural network (CNN) architecture specifically created for the requirements of criminal face identification in order to address these problems. ScanNet leverages advanced CNN techniques, combining Convolutional layers, Separable Convolutions, Batch Normalization, and the Swish activation function to enhance feature extraction and classification. The architecture incorporates Spatial Dropout and Global Average Pooling to manage overfitting and improve generalization, ensuring that the model performs effectively across diverse and unseen images. Additionally, the design includes multiple Dense layers to fine-tune the learned representations, while the SoftMax output layer classifies the images into predefined categories. The model was trained on a dataset of criminal faces spanning 10 unique classes. To ensure robustness, extensive data augmentation techniques were employed, including random rotations, shifts, shearing, zoom, and horizontal flips. These techniques simulate real-world variations in criminal images and improve the model's ability to handle variations in input data. The data pre-processing steps and augmentation strategies were crucial in enhancing the overall performance of ScanNet, especially in dealing with the complexities of criminal identification from facial images. To evaluate the effectiveness of ScanNet, a comprehensive comparison was made with traditional models such as ResNet and VGGNet, widely recognized for their performance in image classification tasks. Additionally, its lightweight design ensures faster training and inference times, further enhancing its suitability for real-world deployments. Further optimization through techniques like transfer learning, hyperparameter tuning, and expanding the dataset to include a broader range of profiles will improve accuracy and adaptability. Moreover, leveraging edge computing technologies can enable low-latency, real-time detection in environments with limited computational resources. The experiments demonstrated that ScanNet consistently outperformed these models, offering improvements in both accuracy and robustness. According to the findings, the suggested architecture offers a more dependable and effective solution for security and law enforcement applications, and it is especially well-suited for criminal face identification. This paper contributes to the growing field of facial recognition technologies by presenting a novel model designed specifically for criminal face detection. The findings emphasize the importance of custom architectures in overcoming the limitations of general-purpose models. By adhering to these principles, ScanNet not only enhances security but also fosters public trust in the responsible use of AI-driven surveillance systems. Furthermore, the model's efficiency in managing large datasets and high-throughput video streams makes it a practical solution for modern surveillance systems. Future work could explore further optimization for real-time detection and expand the dataset to include more diverse criminal profiles, enhancing the system's scalability and applicability in broader security contexts. [ABSTRACT FROM AUTHOR]

Copyright of Grenze International Journal of Engineering & Technology (GIJET) is the property of GRENZE Scientific Society and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)