Reversible data hiding using local edge sensing prediction methods and adaptive thresholds

• This research presented an improved Feng and Fan’s approach, which can enhance prediction and hiding capacity.
• Two threshold values are generated through standard deviation analysis for each block and the statistical distribution of the standard deviation.
• This paper proposed a systematic scheme to determine the thresholds to obtain more accurate prediction results.
• The experimental results demonstrate that a high image quality is achieved.
• The application of the proposed method to complex images provides improved noise and embeddable capacity properties.

This paper proposes an improved reversible data hiding technique by utilizing Lukac, Feng, and Fan׳s expanding method. The scheme uses standard deviation to analyze the complexity of an image to determine the prediction method. The proposed scheme applies two thresholds to control the hiding capacity. The first threshold is used to decide which equation is better to calculate the prediction value. The second threshold is used to estimate the cover pixel is embeddable. The threshold values greatly affect the capacity and image quality of the stego image. Hence this paper develops a systematic scheme to determine the thresholds to obtain more accurate prediction results. A shifting-based histogram technique is then built from the prediction error values and secret message is embedded in pixels with low predicted error value. The experimental results demonstrate that a high image quality is achieved. From the experimental results, we can see that the image quality of the proposed scheme is better than that of Lukac׳s or expanded Lukac׳s method. Especially when the hiding capacity is set to be low, the PSNR value of the proposed is 5 db more than that of the other two methods. The smooth images, such as Lena, Goldhill, Living room images have the smallest difference in PSNR among the algorithms, while the complex images, such as Mandrill, Peppers, Barbara, Boat, and Pirate images show improved PSNR values in the proposed embedding method compared to the other two algorithms. On average, the proposed method can improve 2 dB than Feng׳s method. Specially, for complex images such as Mandrill, the proposed scheme can obtain more than 3 dB gain. The application of the proposed method to complex images provides improved noise and embeddable capacity properties.