Generating facial images for steganography based on IGA and image morphing

In our earlier study, we proposed an image morphing based method for information hiding. The basic idea is to hide a secrete image into a morphed image which is obtained from the secrete image and another reference image. To make this method practically useful, it is necessary to produce natural morphed images. This is a necessary condition conceal the existence of the secret image. To produce natural morphed images, we should choose a proper feature point set (FPS) for morphing. This is a tedious work if we do it manually, because the number of possible FPSs is very large. To solve the problem more efficiently, we adopted interactive genetic algorithm (IGA) in this study and conducted experiments for generating facial images. Results show that, if we provide a relatively good initial FPS, IGA can finetune the FPS, and produce more natural facial images with limited number of evaluations.