Identification of criminal & non-criminal faces using deep learning and optimization of image processing

• Published in: Multimedia tools and applications Vol. 83; no. 16; pp. 47373 - 47395
• Main Authors: Sivanagireddy, K., Jagadeesh, S., Narmada, A.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2024; Springer Nature B.V
• Subjects: Algorithms; Ant colony optimization; Closed circuit television; Comparative studies; Computer Communication Networks; Computer Science; Data collection; Data Structures and Information Theory; Deep learning; Face recognition; Facial recognition technology; Feature extraction; Image processing; Machine learning; Multimedia Information Systems; Principal components analysis; Special Purpose and Application-Based Systems; Surveillance systems; Deep learning; Densenet 169; Face recognition; Face identification; ACO; Classification
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-17471-7

Since identifying criminals is a crucial function of intelligent surveillance systems, it has attracted a lot of attention. Although various approaches are developed for criminal face recognition, they cannot accurately identify the criminal faces. In this study, a novel advanced deep learning model was designed for accurate identification of criminal face from the CCTV images. The developed model utilizes five major phases namely, data collection, pre-processing, feature extraction, feature selection and classification. The study utilizes the data collected from the National Institute of Standards and Technology (NIST) containing criminal and non-criminal face images. The developed model employs Haarcascade algorithm for scaling and transforming the raw images into appropriate format for subsequent analysis. Further, the designed model utilizes Principal Component Analysis (PCA) and Ant Colony Optimization (ACO) for feature extraction and selection, respectively. Finally, the face recognition task was performed using the DenseNet 169 classifier. The developed framework was designed and implemented in Pytorch software and the result metrics are estimated. Furthermore, a comprehensive comparative study was conducted to validate the performances of the developed model with the conventional deep learning models. The experimental results and comparative study illustrate that the designed model outperformed the traditional models.