Convolutional-based variational autoencoders for face privacy protection in video surveillance

Mallepogu Sivalakshmi,K. Rajendra Prasad,C. Shoba Bindu
DOI: https://doi.org/10.47974/jdmsc-1975
2024-01-01
Journal of Discrete Mathematical Sciences and Cryptography
Abstract:The ubiquity of high-quality video surveillance owes much to advancements in imaging technology and data transmission. Presently, exposing an individual’s face in photographs can infringe upon their right to privacy. Real-world face de-identification is a typical task beyond removing private information, considering the specific intent behind image usage. This paper introduces a novel deep learning model (NDLM) designed to safeguard facial privacy in video surveillance, structured around two key phases: face detection and privacy protection. Initially, surveillance video data is collected from online sources. In the initial phase, face detection is achieved through the integration of a Hybrid Convolutional Neural Network (HCNN), which combines a convolutional neural network (CNN) and the improved Fire Hawks algorithm (IFHA) for optimal performance. IFHA assists in ensuring privacy protection. This integration incorporates service quality preservation into the loss function, facilitating the generation of facial images with controlled quality preservation. The approach effectively manages various service quality measures and is adaptable across diverse service contexts. The proposed methodology’s effectiveness is evaluated to demonstrate its efficiency when compared to traditional methods.
What problem does this paper attempt to address?