Face Detection and Encryption for Privacy Preserving in Surveillance Video

A number of techniques have recently been proposed for privacy preserving in video surveillance. Most of them are irreversible or have interference effect to the observation and recognition of human activities. In this paper, we address these issues by developing an effective method including face detection and encryption. In face detection, skin-color based approach fusing with fuzzy clustering is produced to detect facial candidates coarsely, and then we refine face by using SVM classifier. In face encryption, a reversible hybrid encryption (decryption) scheme based on spatial and value scrambling models is proposed. Simulation results verify the proposed mechanism can effectively detect and obscure faces while leaving the activities comprehensible and has high key sensibility for reducing the probability of attacking.