Privacy preserving face recognition in encrypted domain

This paper presents a privacy preserved face recognition framework which combines cryptographic techniques and image processing operations. Unlike other approaches, in this work, face recognition is directly applied on the encrypted data. Cryptographic keys are generated by using a random number generator based on a continuous chaotic system, which is proved to be unpredictable. One key is generated for each pixel of the input image. Every pixel of the input image is gone through a bitwise XOR operation with the corresponding key. The resulting encrypted image is used for face recognition process, where eigenvalues are calculated for Principle Component Analysis based recognition. Two popular databases are used for experiments to measure face recognition accuracy, structural similarity index (SSIM) and PSNR between original and encrypted images. Low PSNR and SSEVI values show dissimilarity level, hence successful preservation of privacy. Face recognition is achieved with accuracy rates close to original recognition process.