report:
[attachment=8389]

Contents
1 Introduction 1
1.1 Organization Of the Report . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Edge Directed Prediction 2
3 Embedding phase 4
3.1 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.1.1 Input the pixel from the original image . . . . . . . . . . . . . . 4
3.1.2 Predict the pixel values . . . . . . . . . . . . . . . . . . . . . . . 4
3.1.3 Compute the dierence . . . . . . . . . . . . . . . . . . . . . . . 5
3.1.4 Embed the secret data . . . . . . . . . . . . . . . . . . . . . . . 5
3.1.5 Output the pixel value . . . . . . . . . . . . . . . . . . . . . . . 5
4 Extracting phase 7
4.1 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.1.1 Input pixels from the stego image . . . . . . . . . . . . . . . . . 7
4.1.2 Predict the pixel values . . . . . . . . . . . . . . . . . . . . . . . 7
4.1.3 Compute the dierence . . . . . . . . . . . . . . . . . . . . . . . 8
4.1.4 Restore the image and retrieve the secret data . . . . . . . . . . 8
4.1.5 Output the pixel values . . . . . . . . . . . . . . . . . . . . . . . 8
5 Evaluation of the proposed scheme 10
5.1 Payload capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5.2 Stego-image quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
6 Conclusion 12
References 12

Abstract
Any kind of distortion is intolerable in the eld of sensitive images, such as medical or
military imaging. Reversible or loseless data embedding schemes are the only solutions
available there. In the proposed scheme, a sender embeds invisible information into
an image bit by bit. A bit is embedded at a time based on the values of predicted
and original pixel values. Since this process is reversible, the reciever can extract the
embedded data and restore the original image later. While preserving the quality of
the stego-image, this method can provide greater payload capacity and data hiding
capacity.

Chapter 1
Introduction
Data embedding techniques are extensively used in copyright marking and in the eld
of steganography. In copyright marking, a logo or secret information is embedded into
an image by the owner using any of the available data embedding techniques. Later,
this secret information is retrieved by the reciever for authentication. In steganogra-
phy, secret data is hidden in a cover image without being suspected by attackers.
The Edge Directed Prediction (EDP) scheme generates a prediction pixel value
based on prediction coecients and past casual neighbours. If the dierence between
the predicted and original pixel value is larger than a predetermined threshold, we
hide a secret bit in that pixel according to our proposed modiction strategy. A stego
image is generated after completing the embedding phase. In the extraction phase,
original pixel value is restored after extracting the secret bit.
1.1 Organization Of the Report
1. Chapter 2 introduces Edge Directed Prediction for loseless compression.
2. Chapter 3 describes the embedding phase.
3. Chapter 4 describes the extraction phase.
4. Chapter 5 provides the evaluation of the above scheme.

Chapter 2
Edge Directed Prediction
In Li and Orchards scheme of edge prediction of lossless images, based on pixel lo-
cations in the original image, three kinds of predictors, median edge detector , low
and high-order EDPs, are used to predict pixels in an image so that serial compres-
sion codes can be generated. Later, these serial compression codes can be used to
reconstruct an original image. Pixels must be predicted pixel by pixel when they are
located in the MED and low-order EDP areas. Otherwise, they are predicted edge-by-
edge. The whole process is adapted according to the predened threshold. A bitmap
is required to record which pixel is used to start the switching strategy after high-
order EDP prediction. To achieve lossless compression, the prediction error between
the original pixel and the predicted one also must be recorded for later receivers to
losslessly reconstruct the image.