Feeling awareness in interactive differential evolution based facial image generation

In our earlier studies, we have proposed an image morphing based steganography. The basic idea is to embed a secrete image into a stego image through morphing. The stego image is a morphed image obtained from the secrete image and some other reference images. To make this method more practically useful, it is essential to make the stego image as natural as possible. For this purpose, we should define a proper feature point set (FPS) for morphing. To find a good FPS efficiently, we use interactive differential evolution (IDE) in this paper. One problem in using IDE is that, in the evolution process the human user may change his/her evaluation criterion subconsciously. This is not desirable for finding an objective evaluation function based on the generated images. To solve this problem, we investigate the effectiveness of k-NN based feeling awareness, and propose a reminder for stabilizing the “user feeling” in the evolution process. Experimental results show that the reminder is useful for reducing the chance of criterion shift. Results also show that increasing k alone may not improve the performance of the reminder for user feeling modeling.