Figure 1. Adaptive Region of an Image Based on Color
Transmitting the secret information by Steganography plays a vital role in Human Visual System (HVS). The carrier media such as image or audio or video can be used to hide the information. Steganalysis is a technique used to get rid of cheating, by identifying the hidden information from the carrier media. The identification of embedded information or message from the carrier media produces higher success rate to steganography methods. Image Steganography is the art of hiding the message or a file or an image by taking the image as carrier media. Based on the adaptable regions, the content is hidden and this method is termed as Adaptive image steganography. Dealing with retrieval of embedded content from the adaptable region of cover image is known to be Adaptive Image Steganalysis. The Blind Steganalysis is the ability to attack the stego image without the knowledge about steganography. Its Counter method, attacks the stego image by significant method used for steganography. In existing method, Enhanced canny edge detector is used to extract the features of the image better than other edge detectors, but smoothens the boundaries including noise and fails to identify the false edges. In the proposed method, Watershed method is used to segment the Adaptive regions from the stego image. The Markov Random Fields (MRF) extracts the features from the segmented adaptive region. The precision and recall is calculated after identifying the adaptive region with its payload location and hidden content using SVM (Support Vector Machine) classifier. An SVM is a binary classifier, classifies data by finding the best hyperplane which separates all data points of one class from the other class. After the classification and identification of payload location, the message is extracted from the hidden region by reversible two LSB (Least Significant Bit) bits.
Steganography is the practice of concealing or hiding a file, message, image, or video within another file, message, image, or video. The advantages of Steganography is by hiding the content in the image or carrier or cover media the information's can be transferred with security and safety. Retrieving the hidden message from the stego image which acts as a cover media or carrier is called as Image Steganalysis. Hiding the information from vision of human is the role played in Adaptive steganography. The stego images have the adaptive regions and it is the sub region of the image whose intensities are closer with neighboring pixels. This intensity level in the adaptive region helps in hiding the message in a colored image and increases the success rate in steganography. The Steganalysis method that is deliberated for gray scale images and can also be used for color images by considering the color images in three times of greater gray scale images[1].
Based on adaptive regions, Multimedia signals were embedded in color images which cheat the visual systems of human[2]. The Blind Steganalysis is the ability to attack the stego image without the knowledge about steganography. Blind Steganalysis is a merge of feature extraction and feature classification[3]. Before finding the hidden content or message, the image is to be segmented to find the region in which the content is hidden. The segmentation of Adaptive regions from the stego image is done by using Watershed method. The Markov Random Fields (MRF) extracts the features from the segmented adaptive region of the stego image. Using the extracted information from the adaptive region by MRF, the hidden information is found by calculating the precision and recall value. After identifying the adaptive region with its payload location and hidden content using SVM (Support Vector Machine) classifier, the precision and recall is calculated. An SVM is a binary classifier that classifies data by finding the best hyperplane, which separates all data points of one class from the other class. After the classification by the SVM classifier and identification of payload location, the information is extracted from the hidden region by reversible two LSB (Least Significant Bit) bits. The methodologies used are Watershed method to segment the image, Markov Random Fields (MRF) is used to extract the feature from adaptive region, Support Vector Machine (SVM) for finding the hidden information.
The main objective is to find the hidden information from the stego image by identifying the adaptive regions using Watershed Segmentation method. The methodologies used are Watershed method, Markov Random Fields (MRF), and Support Vector Machine (SVM) for finding the hidden region and information.
Lamia Jaafar Belaid and Walid Mourou[4], used on approach based on the watershed transformation. In order to avoid over segmentation, they propose to adapt the topological gradient method. The watershed transformation combined with a fast algorithm based on the topological gradient approach gives good results. The numerical tests obtained illustrate the efficiency of the approach for image segmentation.
E. Punarselvam et al. [5], have worked with the watershed algorithm used to detect the boundaries and edges in images. There are internal and external markers and it has global thresholding, region merging, and splitting or watershed transform.
Jos B.T.M. Roerdink and Arnold Meijster[6], used watershed transform, the method of choice for image segmentation in the field of mathematical morphology. They presented a critical review of several definitions of the watershed transform and the associated sequential algorithms, and discussed various issues which often cause confusion in the literature. The need to distinguish between definition, algorithm specification and algorithm implementation is pointed out. Various examples are given in this paper which illustrates differences between watershed transforms based on different definitions and/or implementations.
Tara Saikumar et al. [7], proposed a new method for image segmentation which combines the watershed transform, FCM, and level set method. The watershed transform is first used to presegment the image so as to get the initial partition of it. Some useful information of the primitive regions and boundaries are obtained.
Ehsan et al[8], . proposed the color space intensity to find the region, where the content should be embedded. Intensity based edge detection was done optimally by canny edge detection algorithm. It read the color images and split it into three unique color channels, then canny edge detector was applied on each channel to finish the edge map. At last, the individual edge maps were grouped into a complete edge map for better results.
In the proposed work, the image segmentation method is used to segment the particular region from the entire image. The Watershed method is the segmentation method used to segment the Adaptive regions from the stego image. The Markov Random Fields (MRF) is the feature extraction method, which extracts the features from the segmented adaptive region. MRFs request to maximize the probability of identifying a labeling scheme given as a particular set of features are detected in the image.
The precision and recall is calculated after identifying the adaptive region with its payload location and hidden content using SVM (Support Vector Machine) classifier.
The SVM classifier is a best classifier for binary classification of image into a cover or stego image. The Payload locations in an image are identified using reversible process of LSB (Least Significant Technique) steganography[9]. Each pixel value is compared for identifying the payload locations of adaptive region for better precision. After the classification and identification of payload location, the message is extracted from the hidden region by reversible two LSB (Least Significant Bit) bits. Thus the proposed method is better than the simple Enhanced canny edge detection operator in finding the adaptive region for blind image Steganalysis along with the Support Vector Machine (SVM) Classifier.
Adaptive region of an image is the sub region of the image, where the intensity level of pixel of the region is closer to the neighboring pixel. This helps steganography for hiding the information in adaptive region. The process of identifying the adaptive region from where the hidden information should be retrieved is termed as Adaptive Steganalysis. Intensity level of the image plays a vital role for finding the adaptive region.
Human Visual System is not aware of the presence of the hidden information[10]. Adaptive region identification is done using watershed method which is an segmentation. The segmentation is done to the stego images. The Markov Random Fields (MRF) extracts the features from the segmented adaptive region.
The precision and recall is calculated after identifying the adaptive region by segmentation and feature extraction, with its payload location and hidden content using SVM (Support Vector Machine) classifier [11]. After the classification and identification of payload location, the message is extracted from the hidden region by reversible two LSB (Least Significant Bit) bits and the message length estimation[12]-[16]. Embedding the message in adaptive region, results in higher success rate of steganography[17]. Adaptive regions are identified for retrieval of hidden information, which is shown in Figure 1.
Figure 1. Adaptive Region of an Image Based on Color
The block diagram for Adaptive Steganalysis for LSB technique is shown in Figure 2. The proposed method aims in identification of the adaptive region using the segmentation method. The segmentation method used is Watershed method considers the stego images and identifies the adaptive region. The features from the segmented image are extracted by using Markov Random Fields (MRF). Support Vector Machine classifier calculates the precision and recall after finding the adaptive region and the hidden information is extracted by reversible two LSB from the adaptive region.
Figure 2. Block Diagram for Adaptive Steganalysis for LSB Embedding
The following algorithm discuss about the adaptive Steganalysis for LSB embedding.
5.1.1 Algorithm 1: Adaptive Steganalysis for LSB Embedding
1. Consider the set of images.
2. Apply Watershed segmentation method to find the adaptive regions of the image.
3. Consider the adaptive regions that are identified, and Markov Random Fields are applied to extract the features.
4. Support Vector Machine classifier classifies the image into stego or cover image.
5. The hidden information is extracted by reversible two LSB from the adaptive region of stego image.
6. Estimate the length of the hidden message from stego image by identifying payload location.
7. Calculate the precision and recall after finding the hidden message from stego image.
8. Extract the message.
9. Return the hidden message.
The Watershed method segments the image. One aim of this watershed segmentation is to show how the use of mathematical morphology operators are useful in image segmentation. Figure 3 shows the steps followed to perform watershed segmentation. Initially, the input image is converted to gray scale image. The Gradient image is found from the gray scale image and reconstruction of image is done by using morphological opening. Then, mark the foreground and background objects which calculate the regional maxima of reconstructed image to obtain good foreground markers. Regional maxima are connected components of pixels with a constant intensity value t, whose external boundary pixels all have a value less than t. Modify the gradient image using foreground and background markers. Modify an image so that it has a regional minima only in certain desired locations. Now, compute watershed transform of the modified gradient image to get segmented output. The minima at level i + 1 are given by,
Figure 3. Steps for Watershed Segmentation
Markov Random Fields have strong mathematical foundation and ability to provide global optima even when defined on local features proved to be the foundation for novel research in the domain of image analysis, de-noising and segmentation. MRFs are completely characterized by their prior probability distributions, marginal probability distributions, smoothing constraint as well as criterion for updating values. The criterion for image segmentation using MRFs is restated as finding the labeling scheme which has maximum probability for a given set of features. The broad categories of image segmentation using MRFs are supervised and unsupervised segmentation. In terms of image segmentation, the function that MRFs seek to maximize is the probability of identifying a labeling scheme given a particular set of features is detected in the image. This is a restatement of the Maximum a posteriori estimation method.
1. Define the neighborhood of each feature (random variable in MRF terms). Generally, this includes 1st order or 2nd order neighbors.
2. Set initial probabilities i for each feature as 0 or 1, where i is the set containing features extracted for pixel and define an initial set of clusters.
3. Using the training data compute the mean (li) and variance (li) for each label. This is termed as class statistics.
4. Compute the marginal distribution for the given labeling scheme using Bayes' theorem and the class statistics calculated earlier. A Gaussian model is used for the marginal distribution.
5. Calculate the probability of each class label, given the neighborhood defined previously. Clique potentials are used to model the social impact in labeling.
6. Iterate over new prior probabilities and redefine clusters such that these probabilities are maximized. This is done using a variety of optimization algorithms.
7. Stop when probability is maximized and labeling scheme does not change. The calculations can be implemented in log Likelihood terms as well.
Support Vector Machine (SVM) classifier is a binary classifier which is classified as linear and non linear SVM. Support Vector Machine is the best classifier for image classification. The image classification is done based on the following steps.
1. Consider the image dataset.
2. Preprocess the images in the dataset.
3. Classify the dataset into training dataset and test dataset.
4. The input images are trained to identify the stego images and the cover image based on its adaptable region.
5. The classification is done on testing phase to separate stego images from cover image.
The Watershed segmentation algorithm performs better segmentation for identifying the adaptive region. In the adaptive region, LSB reversing for information retrieval is used so that fast and efficient information retrieval is done. Figure 4 shows the segmentation for adaptive region using Watershed segmentation.
Figure 4. Adaptive Region Identification using Watershed Segmentation
The performance measure of the proposed algorithm is calculated using the identification of adaptive region so that the LSB reversing can be concentrated on a particular region which enhances the performance of Steganalysis. Precision (Positive Predictive Value) is used to identify the detection rate of hidden information from the stego images based on adaptive region identified using Watershed segmentation algorithm. The Positive Predictive value is calculated by using Equation (2), Recall is the retrieved fraction of relevant instance. The Recall value is calculated using Equation (3).
Where, P is the Positive Predictive Value (Precision), FP is the False Positive, and TP is the True Positive. Precision is calculated based on two metrics called True Positive and False Positive. R is Recall, FN is False Negative and TP is True Positive. Recall is calculated based on two metrics called True Positive and False Negative. Table 1 shows the Precision and Recall values for 600 images of Enhanced Canny Edge Detection method and Watershed segmentation method with different Embedding Ratio by using TP, TN, FP and FN of the images.
Table 1. Comparison of Precision and Recall Value of Enhanced Canny Edge Detector and Watershed Segmentation Method
Comparing to the Enhanced canny edge detector, the watershed segmentation method is better in embedding the data or information because the precision value of the Enhanced Canny Edge detector is less when compared to Watershed segmentation method. True Positive (TP) defined as the hidden information in the input image is identified correctly as stego image. True Negative (TN) states that the input image has hidden information, but it is classified as Cover image with no hidden information. False Positive (FP) states there is no hidden information in the input image, but it is classified as stego image. False Negative (FN) identifies the cover image correctly, if the input image is a cover image. In Enhanced Canny operator, the TP value is less when compared to the Watershed segmentation method.
Figure 5 represents the graphical representation of the True Positive, True Negative, False Positive, False Negative, Precision, and Recall value of both Enhanced Canny edge detector and Watershed method.
Figure 5. Graphical Representation of Precision and Recall Values
The graph implies that, Watershed segmentation is better in identifying adaptive region in effective and faster rate. Reversing two LSB Embedding methods determines the payload location of jpeg images with Positive Predictive value. The proposed method works better in identifying payload location from an adaptive region. In the proposed method, the watershed method is used to segment the region and MRF extracts the features of the stego image. SVM classifier classifies the images as stego or cover images. Experimental analysis shows that Watershed segmentation method is an effective method which is used to detect the adaptive regions. When the embedding ratio of the image increases, the precision value for Watershed method is best when compared to the Enhanced canny operator.