Prediction error expansion (PEE) based reversible data hiding (RDH) requires a good predictor for predicting the pixel. The Prediction Error ( \(PR_{Err}\) ) is used to insert the secret data into the original cover image pixels. Various pixel prediction algorithms exist in the literature for better predicting the cover image pixel. Several scholars have put forth prediction methods that use the different gradient estimations. This gradient-based pixel prediction strategy is investigated in more detail in this paper. Various neighborhood contexts surrounding the current pixel have been investigated for better exploring gradient estimations. Experiments have been reported to test the size of the neighborhood being considered to estimate the gradient. Two strategies for selecting directions based on gradient magnitudes are also explored. A novel adaptive embedding strategy has been proposed for embedding the data into the original pixels. Experimental results lead towards an improved gradient based prediction using adaptive embedding strategy in the context of reversible data hiding
