New Informed Non-Blind Medical Image Watermarking Based on Local Binary Pattern

Medical image watermarking represents a promising alternative tool regarding security, digital rights, authenticity and integrity issues. In this paper, an informed medical image watermarking scheme is proposed based on local binary pattern LBP. A watermark is built based on the significant information extracted from the host image by LBP and will be addressed to be embedded through a linear interpolation. Scenarios of geometric and non-geometric attacks have been realized on the watermarked images to evaluate the robustness of the embedded watermark in the extraction process. Furthermore, it is verified through achieved experiments that the proposed scheme is imperceptible and more robust from the achieved results which are very encouraging.


Introduction
The digital technologies, the explosion of communication networks and the ever-growing enthusiasm of the general public for new information technologies are leading to an increased traffic of multimedia (images, videos, texts, sounds, etc.). The extent of this phenomenon is such that essential concerns now arise regarding the protection and control of the exchanged data. Indeed, by their digital nature, multimedia documents can be duplicated, modified, transformed and distributed easily. Under these conditions, it becomes therefore necessary to develop systems to enforce copyright, protecting the integrity of documents and authentication [1]. In this context, digital watermarking very quickly appeared as the alternative solution to reinforce the security of multimedia contents.
The main idea of image digital watermarking is to hide in digital image subliminal information (i.e. invisible or imperceptible) to ensure a security level (copyright, integrity, authenticity purposes, etc.). One of the specificities of digital watermarking compared to other techniques, such as for example simple information storage in the header of file, is that the watermark is intimately to data. Therefore, the watermarking is theoretically independent from the format of file and can be detected or extracted even if the image has undergone modifications or is incomplete.
Watermark embedding techniques can be classified into two main categories depending on the targeted objective: the watermarking domain and the type of watermarking. For the watermarking domain, two essential domains are distinguished: 1) the spatial domain in which directly modifies the pixels without any preliminary processing [2 , 3 ] to embed the watermark; and 2) the frequency domain [4 , 5 ]which requires transforms before embedding the watermark such as the Discrete Cosine Transform [6 , 7 ], Discrete Wavelet Transform [8 , 9 ], Singular Value Decomposition [10,11 ], etc. For the second category which consists of the type of watermarking, we define three types: Blind watermarking [12] in which we need only the watermarking key to extract the attacked watermark; semi-blind watermarking [13] where the host image is required to extract the attacked watermark and finally the non-blind watermarking [14] while the original watermark is needed to extract the attacked watermark.
A watermarking scheme must be robust against different scenarios of geometric and/or nongeometric attacks which mean that the watermark can be extracted even if the watermarked image is attacked. Similarly, the computational time parameter should not be neglected, especially with the growth of data, which means that this parameter must also be taken into consideration for the watermarking scheme to be applicable in real time.
We present through this paper, a new medical watermarking scheme both for embedding and extracting watermark. The proposed scheme is based on local binary pattern LBP to be involved to build an informed watermarks based on the LBP operators. A different scenario of attacks has been applied on the watermarked images to evaluate the robustness of the embedded watermarks. The obtained results are well discussed and evaluated.

Related work
Authors in [15] propose a watermarking approach with a blind manner which consist to incrust watermark directly and without any transform to the blue component in the RGB image. The approach exploits the DC coefficient to realize embedding watermark in different regions of the host image. The extraction process is based on the same way when embedding watermark on the watermarked image and the key-based quantization. Authors show that the proposed approach gain an acceptable invisibility factor when embedding watermark and resist against some kind of attacks such cropping, and noise JPEG compression.
In [16], a spatial domain watermarking technique has been proposed. The technique is based on combining the DC behavior in its original value and its value when applying a Fourier transform. Embedding and extracting watermarks is basically based on the changed value of the DC component of each block. Authors illustrates that the proposed technique have shown the invisibility of the watermark on the watermarked images. The technique was tested against a few numbers of attacks such JPEG compression and adding noises. The main contribution of this technique is its low computational time.
In the watermarking scheme proposed in [17], authors suggested to achieve the watermark embedding on specific regions in the host image. The approach is principally based on using two masks which are used to distribute the watermark information to the neighboring pixels in the selected region. The firs mask modulates blue component of the RGB image and has a role to the tune of the watermark bit. While the second mask which is considered as a compensating mask that adjusts red and green color channels.
Authors in [18], proposed a Least Significant Bits (LSB) spatial watermarking approach through an image gradient and chaotic map. The image is divided into non overlapping blocks, and the gradient of each block is calculated. As known, the gradient expresses a good information regarding changes in an image. A chaotic substitution box (S-Box) is used to scramble the watermark according to a piecewise linear chaotic map (PWLCM). The embedding payload introduces a compromise between robustness and watermark invisibility. The tests achieved through this approach show a satisfactory enhancement in term of watermark robustness against geometrical attacks. The approach maintains also an acceptable imperceptibility of the watermark.

Local Binary Pattern
Local Binary Pattern (LBP) is a technique in which many image features can be expressed such local texture (spot, Spot/flat, line end , edge, corner, etc..) operator for a gray-level changes. This technique is obtained from information regarding texture in a local neighbourhood. The LBP operator thresholds is a neighbourhood gray value centre, by presenting the outputs as a sequence of binary code that defines the local texture pattern as shown in equation 1. The LBP performs remarkably with applications needing fast feature extraction and texture classification due to its discriminative power and computational simplicity. The value of LBP code of a pixel ( xc, yc) is given by: (1)  The different steps to calculate the local binary pattern LBP for an image are illustrated in the following algorithm.

LBP operator computation steps
Step 1. Converting the input image into graylevel.
Step 2. Selecting the Q neighborhoods For each pixel (gQ), select the P neighborhoods surrounding the central pixel.
Step 3. set center pixel (gc) as a threshold for its Q neighbors.
Step 4. Set to 1 if the adjacent pixel value is >= to the value of the center pixel, 0 else.
Step 5. Calculate the LBP operator with a sequentially and counterclockwise manner, write a binary number consisting of digits adjacent to the center pixel according to equation 1. For a given gray-level image, we illustrate how to compute an LBP operator for a given block of size 3×3 as shown in figure 2. Through LBP we can define many local primitives of an image such curves, spots, edges, etc…

Proposed watermarking process
The watermarking scheme that we propose through this paper consists of three essential phases. The first phase concerns watermarking embedding within a host image in an imperceptible way to guarantee its robustness. The second phase consists of applying geometric and non-geometric attacks on the images watermarked. The role of the third phase is to extract the watermark after applying attacks. The images used in our tests are illustrated in the figure 5 and are of size 255×255 and in gray-level [19]. We detail the three steps in the following.

Spot
Spot/flat Line end Edge Corner Figure 5. the used images in tests

4.1.Watermark embedding step
To guarantee the robustness of the watermark against different kinds of attacks, the watermark must imperatively be invisible (imperceptible). The imperceptibility is manipulated by a linear interpolation as shown in equation 2. Whether the parameter α closer to 1 more the watermark is invisible. More α is close to 0, more the watermark becomes visible. Figure 6 illustrates the visibility/invisibility of the watermark by changing the values of the watermarking key α.

= (1 − ) +
(2) Each medical image has been embedded by its corresponding LBP image. Note that the LBP matrix is calculated with an overlapping blocks of size 3×3.

4.2.Scenario of attacks
We tested the proposed approach on a database of 25 images in gray-level and of size 255×255.The watermarked images were a subject of many attacks through Stirmark benchmark  Figure 7. Some of applied attacks on the watermarked images As mentioned previously, geometric and non-geometric attacks as shown in figure 7 have been applied to different watermarked images such: chest, heart, shoulder, ORT and knee. We summarize here some attacks like JPEG compression, Median filtering, rotating with a chosen angle, adding noise and removing lines (vertically and horizontally). Attacking the watermarked images will be useful in the next section (watermark extraction) in order to decide about the watermark resistance against such attacks.

4.3.Watermark extraction step
In this step, we try to extract the attacked watermark wa from the attacked watermarked image iwa. This process is the reverse operation of watermark embedding. The extraction of the attacked watermark is achieved in a non-blinded way as defined in equation 3.
The general watermarking scheme covering embedding, attacks and extracting watermark is shown in figure 8.

Figure 8. General watermarking scheme
We extracted the different watermarks from their corresponding attacked watermarked images against every attack as shown in figure 9. The extracted watermarks will be a subject of discussion in the next section regarding their robustness.

Results and discussion
The evaluation of the results obtained will be defined on the calculation of three widely used metrics in image watermarking: Peak Signal Noise Ratio (PSNR), Correlation coefficient (CC) and the Bit Error Ratio (BER).
PSNR [21] is a logarithmic function of Mean Square Error (MSE) interpreted as a corrected version of the Signal-to-Noise Ratio. A PSNR value that exceeds 34db means a similarity between two images. A high similarity between two given images is expressed when →∞.
The PSNR is calculated using equation 4: Where i(p,q) and iw(p,q) are the pixels (p,q) in the original image i and the watermarked image iw respectively. × is the size of the image.

Host images i
Corresponding watermarked images iw  This means that host images and their corresponding watermarked images are visually very similar and the difference cannot be detected by the human visual system. From the PSNR values we can confirm that the approach is imperceptible as shown in figure 10. Now, to evaluate the robustness of the proposed approach, we rely on the correlation coefficients and the Bit Error Ratios between the original watermarks and their corresponding extracted watermarks.
Correlation Coefficient CC [21] represents the similarity between the original image and the attacked one. This coefficient is ranged in [1,-1]; if CC=1 this means that two images are highly identical, if CC=0 this means that two images are completely different. This metric is computed according to equation 5: Where , are intensities of pixel ( , j) in the original watermark image and the extracted watermark respectively. The images w , a are the means intensities of respectively the watermark image and the extracted watermark .
BER [21] (Bit Error Rate) is the quotient of erroneous attacked image bits on the total number of original image bits. It is expressed in percentage as defined in equation 6.   Table 2. BER values between the original watermarks and the corresponding extracted ones.
watermarking field. The approach is purely informed in the sense that the watermark is built from significant features of the host image. This feature is called Local Binary Pattern LBP. The approach was evaluated by applying different kind of attacked on the watermarked images to extract the attacked watermark which help to decide on the robustness of the proposed approach. The evaluation of the obtained results was achieved through well-known metrics in such Correlation coefficients CC and Bit Error Ratio BER. The results obtained are very encouraging and confirm the robustness of our approach.

Statement and Declaration I) Ethical approval
Hereby, I insert Lamri Laouamer consciously assure that for the manuscript "New Informed Non-Blind Medical Image Watermarking Based on Local Binary Pattern" the following is fulfilled: 1) This material is the authors' own original work, which has not been previously published elsewhere.
2) The paper is not currently being considered for publication elsewhere.
3) The paper reflects the authors' own research and analysis in a truthful and complete manner.
4) The paper properly credits the meaningful contributions.

5)
The results are appropriately placed in the context of prior and existing research. 6) All sources used are properly disclosed.
7) The author Take public responsibility for its content.