I-Privacy Photo - Face Recognition and Filtering.

The ever-increasing popularity of Online Social Networks (OSNs) sites for posting and sharing photos and videos has led to unprecedented concerns on privacy violation. The available Online social networking (OSNs) sites offer a limited degree of privacy protection solutions. Most of the solutions focus on conditional access control meaning, allowing users to control who can access the shared photos and videos. This research study attempts to address this issue and study the scenario when a user shares a photo and video containing individuals other than himself/herself (public-level photos and videos). For privacy-preserving, the proposed system intends to support an automated human face recognition and filtering for public-level photos and videos. Our proposed approach takes into account the content of a photo and makes use of face filtering as a strategy to increase privacy while still allowing users to share photos. First, the proposed system automatically identifies a person face frame from a digital image or video. Next, it compares the detected face features to each face vectors stored in the application database. After face recognition step completed, the proposed system filters all un-known persons in the image. Conventual Neural Network (CNN) has been used for face detection step, while deep learning facial embedding algorithms has been used for the recognition. Both have shown high accuracy results in addition to the capability of being executed in real-time. For face filtering, Gaussian algorithm has been used for face blurring as it has been considered a very fast real-time algorithm which allow the user to control the blurring degree. Based on the obtained results after testing the system using three different datasets, we can conclude that our system can detect and recognize the faces in photos and videos using the improved Conventual Neural Network (CNN) for face detection with 91.3% accuracy and K-Nearest Neighbor (KNN) for the face recognition with 96.154% accuracy using I-Privacy dataset.