A Low Distortion Reversible Data Hiding Technique based on Prediction Difference Expansion

Difference expansion based data hiding techniques provide relatively high data hiding capacity. However, increasing the data hiding capacity introduces more distortion to the cover image. A solution to this problem is the use of a distortion control parameter (T) that limits the distortion to cover image. Unfortunately, this will decrease the data hiding capacity. Another factor that affects both data hiding capacity and distortion is the predictor utilized to predict pixel values. This paper presents a new prediction technique that reduces distortion to cover image. In contrast to common predictors of difference expansion techniques that utilize one or two adjacent pixels for prediction, the proposed technique utilizes three or four adjacent pixels. This technique uses about one-fourth of cover image pixels as reference pixels without any distortion and thus, reducing the cover image distortion and resulting in high-quality stego images. The proposed technique achieved average peak signal to noise ratio (PSNR) values in the range 33.29-49.7 dB. The average embedding rate in bits per pixel (bpp) ranged from 0.09 to 0.59 bpp. Moreover, the proposed technique showed to be reversible and does not require a location map.