Stego image quality and the reliability of PSNR

Abstract

Digital image steganography is the art of hiding information in other digital images. Moreover, image quality evaluation has many difficulties such as the amount of degradation or distortion induced in the reconstructed image. The peak signal-to-noise ratio (PSNR) is the most common metric used to evaluate the stego image quality. However, subjective evaluation is the most reliable method to measure the image quality. Therefore, we try to give an answer to the following question: “does the PSNR value of a stego image reflect its actual quality?”. However, JPEG steganography represents a distortion source in addition to the image compression. Therefore, this paper investigates the relationship, if there any, between the PSNR and the subjective quality of stego images. Four steganography methods and five grayscale images are used in this paper. Moreover, an adapted double stimulus continuous quality scale (DSCQS) method has been adopted. As a result, PSNR can not be reliably used because it has poor correlation with the mean opinion score (MOS). Moreover, conclusions derived from only PSNR values of different stego images are quite different from that derived from the MOS values. Additionally, MOS shows that a particular steganography method modifies different test images quality in different ways.