A method of embedding and scrambling external data into digital image is described. Reference points (R) are determined, said reference points have one pixel between each reference point horizontally, vertically and diagonally. Pixels next to reference points are predicted and embedded. The value of pixels next to reference points are predicted with proposed formula for average of least gradient change between neighbouring pixels. The predicted pixels shall be within an acceptable error threshold (?) before it is identified as pixel to be replaced (PE). The reference points (R) can be further sub-sampled to gain sub-set of reference points. Sub-sampling is performed to further degrade the image to hide extra data. A visual quality index can be used as guide for level of image distortion. The method can be operated in loss or lossless mode.