The reversible data hiding in encrypted images by MSB prediction of P. Puteaux and W. Puesch not only provides high embedding bit-rates, but also entails a very low mathematical complexity. This correspondence investigates its security and shows flaws in embedding imperceptibility, unauthorized detection/removal of embedded data and unauthorized access to image content. Secure solutions are discussed.

Pixel-value-ordering (PVO) appears as an efficient solution for high-fidelity reversible data hiding. State-of-the-art PVO schemes split the host image into blocks, sort pixels within the blocks based on their graylevel and, finally, embed data into some differences between the sorted values. This paper investigates the use of the prediction error instead of the pixel graylevel both for ordering and embedding. The proposed prediction-error-ordering (PEO) scheme also introduces a two-stage procedure by splitting each block in two sets following a chessboard pattern and by processing set by set each block. The pixels of one set are used to classify and predict the ones of the other set and vice-versa. The proposed PEO approach outperforms the state-of-the-art PVO schemes.

This paper proposes a new vacating room after encryption reversible data hiding scheme. Hidden data is embedded into the encrypted host image by bit-flipping a preselected bitplane of a randomly formed pixel group. The major novelty of the paper is the use of multiple predictors in an adaptive procedure for detecting between original and modified pixels. Four predictors are used on a context of four neighbors, namely the average of the four pixels, a weighted average based on local gradients, the median and the midpoint. Experimental results are provided. Compared with the state-of-the-art reserving room after encryption schemes, the proposed approach provides higher embedding bit-rates at lower distortion.

This paper proposes a new vacating room after encryption reversible data hiding scheme developed for color images. The proposed scheme uses standard exclusive-or encryption and inherits the main features of vacating room after encryption schemes, namely joint and separate methods for data embedding. The proposed scheme exploits both the correlation between neighboring pixels and the correlation between color channels by predicting the original pixel values on color channel differences. The experimental results show that the proposed scheme can eliminate the main drawback of the vacating room after encryption framework, namely the large embedding distortions.

Pixel-value-ordering (PVO) appears as an efficient technique for high-fidelity reversible data hiding. This paper proposes a reversible data hiding scheme based on the pairwise PVO framework with improved difference equations. Both the pixel pair selection and the embedding algorithms are also streamlined. The proposed scheme uses a block classification approach based on a local complexity metric. Uniform blocks are processed using the proposed improved pairwise PVO algorithm. Slightly noisy blocks are embedded using a classic PVO scheme and noisy blocks are kept unchanged. The optimal embedding parameters for a given capacity are determined by linear programming. The proposed pairwise PVO approach outperforms other state-of-the-art schemes.

This letter revisits the pairwise reversible watermarking scheme of Ou et al., 2013. An adaptive pixel pairing that considers only pixels with similar prediction errors is introduced. This adaptive approach provides an increased number of pixel pairs where both pixels are embedded and decreases the number of shifted pixels. The adaptive pairwise reversible watermarking outperforms the state-of-the-art low embedding bit-rate schemes proposed so far.

The use of local prediction in difference expansion reversible watermarking provides very good results, but at the cost of computing for each pixel a least square predictor in a square block centered on the pixel. This correspondence investigates the reduction of the mathematical complexity by computing distinct predictors not for pixels, but for groups of pixels. The same predictors are recovered at detection. Experimental results for the case of prediction on the rhombus defined by the four horizontal and vertical neighbors are provided. It is shown that by computing a predictor for a pair of pixels, the computational cost is halved without any loss in performance. A small loss appears for groups of three and four pixels with the advantage of reducing the mathematical complexity to a third and a fourth, respectively.

This paper proposes a novel predictor, EGBSW (Extended Gradient Based Selective Weighting), and investigates its usefulness in difference expansion reversible watermarking. EGBSW is inspired by GBSW, a causal predictor previously used in lossless image compression and known to outperform well-known predictors as the median edge detector (MED) or the gradient-adjusted predictor (GAP). The proposed predictor operates on a larger prediction context than the one of GBSW, namely a rectangular window of 16 pixels located around the pixel to be predicted. Similar to GBSW, the extended predictor computes the gradients on horizontal, vertical and diagonal directions and selects the smallest two gradients. Opposite to the classical predictor, EGSBW uses a set of four simple linear predictors associated with the four principal directions and computes the output value as a weighted sum between the predicted values corresponding to the selected gradients. The reversible watermarking scheme based on EGBSW appears to outperform not only the ones based on GBSW, MED or GAP, but also some recently proposed schemes based on the average on the rhombus context. Experimental results are provided.

This paper investigates the use of local prediction in difference expansion reversible watermarking. For each pixel, a least square predictor is computed on a square block centered on the pixel and the corresponding prediction error is expanded. The same predictor is recovered at detection without any additional information. The proposed local prediction is general and it applies regardless of the predictor order or the prediction context. For the particular cases of least square predictors with the same context as the median edge detector, gradient-adjusted predictor or the simple rhombus neighborhood, the local prediction-based reversible watermarking clearly outperforms the state-of-the-art schemes based on the classical counterparts. Experimental results are provided.

The embedding not only into the current pixel, but also into the prediction context was recently proposed as an improvement of difference expansion reversible watermarking algorithms. So far it was shown that the effect of splitting the data between the current pixel and the prediction context decreases the embedding distortion, but increases the prediction error. This paper revisits the case of context embedding for the case of pixel prediction on the rhombus composed of the two vertical and the two horizontal neighbors. For this case it appears that the context embedding can be used not only to reduce the embedding distortion, but also to improve the prediction. The gain provided by the improvement of the prediction outperforms the one provided by the reduction of the embedding distortion. Experimental results are provided.