Experiments on Partial Encryption Performance for Image Sets

This paper reports our experiments on the effectiveness of partial encryption for image sets. Professional image databases often contain a large number of similar images. This fact provides a unique opportunity for partial encryption coding. When the image set is stored as a representative image (the average) and the differences between the originals and the average, one could either encrypt a small percentage of every image in the set, or encrypt the average image partially and a tiny portion of each difference image. In our experiment, the portion to be encrypted is simply hidden from the reconstruction algorithm. No actual encryption algorithm is used. The performance of encrypting (hiding) different portions of the image set is evaluated by visual inspection and by the PSNRs between the original and the reconstructed images. Our algorithm is presented in detail and a Webcam image set and an ultrasound image set are used in the test. Hiding the average and portions of the differences provides a better encryption performance with a small fraction of the computation cost, compared to partial encryption of all original images