International Journal of Emerging Technology and Advanced Engineering Website: www.ijetae.com (ISSN 2250-2459, Volume 2, Issue 1, January 2012)
Reversible Watermarking Techniques for Medical Images with ROI-Temper Detection and Recovery - A Survey 1
Amrinder Singh Brar, 2Mandeep Kaur
1
Master of engineering,IT,U.I.E.T,Panjab university,Chandigarh 2 Assistent Professor,IT,U.I.E.T,Panjab University,Chandigarh 1
[email protected] 2
[email protected]
Abstract-- Modern medical equipments produce mass digital images and data everyday [1]. In addition to diagnosis, these images also serve as valid document evidences in legal trials and insurance claims and as well as education material for illustrations in medical research etc. The integrity of such records needs be protected from unauthorized modification or destruction of information. One of the security measures that can be used is watermarking. It has become an important research area in the field of data security, confidentiality and integrity. Also, in applications like
But with the widespread and increasing use of Internet, these digital images can be easily accessed and manipulated. Considering patient‟s privacy and diagnostic accuracy, the prevention of medical images from tampering tends to be an urgent task [1]. It is required to imbibe the aspects of confidentiality, authentication and integrity with the distribution of these images in the Health Information System. Further, the images should contain patient specific data to be correlated with the appropriate patient; failing which the whole system may be compromised [2].
tele-medicine, confidentiality and integrity of a medical image can be achieved by hiding the Electronic Patient Record (EPR data) in corresponding medical images. However, in the medical images the diagnostic quality is very critical and should not be compromised. The objective of this paper is to discuss some of the most efficient techniques suitable for watermarking of medical images. These methods are reversible in nature, i.e, complete information without distortion can be recovered. The techniques are based on modified difference expansion and CDCS. These techniques aim at increasing the data hiding capacity, without distortion of diagnostically important information. Mechanism for ROI authentication and temper detection is also provided to judge its integrity and fidelity.
Another fact is that in recent years, health care systems involve a large amount of data storage and transmission such as patient information, medical images, electronic patient records (EPR) and graphs [3]. Transmission of such a large amount of data when done separately using ordinary commercial information channels like internet, it results in excessive memory utilization, an increase in transmission cost and time and also make that data accessible to unauthorized personnel [4]. In order to reduce storage and transmission cost, data hiding techniques are used to embed patient information with medical images. These data hiding techniques can also be used for authentication and tamper detection so that integrity of region of interest (ROI) can be judged [5].
Medical image; ROI-based; Tamper Detection; Reversible Watermarking; Histogram modification Key words—
Digital watermark provides three objectives in medical images: data hiding (embedding information to make image useful), integrity control (to verify that image has not been modified without authorization) and authenticity (to verify that image is really what the user supposes it is) [6].
I. INTRODUCTION Healthcare institution that handles a number of patients, opinions is often sought from different experts. It demands the exchange of the medical history of the patient among the experts which includes the clinical images, prescriptions, initial diagnosis etc.
In medical imaging applications, there are stringent constraints on image fidelity that strictly prohibit any permanent image distortion by the watermarking [7].
32
International Journal of Emerging Technology and Advanced Engineering Website: www.ijetae.com (ISSN 2250-2459, Volume 2, Issue 1, January 2012)
II. CLASS DEPENDENT CODING SCHEME For instance, artefacts in a patient's diagnostic image due to image watermarking may cause errors in diagnosis and treatment, which may lead to possible life-threatening consequences [8]. Thus, to overcome the problem of occurrence of artefacts and to produce zero distorted or noise free watermarked medical images, various reversible, fragile watermarking schemes have been introduced by various researchers out of which some are surveyed in this paper. In practice, diagnosis is performed on medical images before being directed to the long-term storage, thus the significant part of the image is already been determined by doctors involved in the diagnostic process. The significant part is called ROI (Region of Interest). As information in medical images should not be modified, the watermark is usually being embedded in the RONI (Region of Non Interest) as this region is not important for process of diagnosis [3].
Text Processing Phase r, n
EPR
Image Processing Phase Hash on ROI
CDCS, Redundancy and Interleaving
Medical Image
ROI x1,y1,x2 ,y2
RONI 8x8 Block Division Calculation of robust difference parameter „α‟. ∑ Stego Image
Figure 1.Proposed System
Reversible watermarking has been proposed as a way to integrate multimedia data, and several medical image authentication schemes based on this type of watermarking have been proposed to validate the integrity of medical images. Some of these schemes use spatial domain to embed watermark bits into LSB of grayscale values of pixels in the host image [9] while others use transform domain and embed Watermark bits into coefficients of transformed image [10]. Most of the existing medical image authentication schemes use a hash function to create a digest of the image and embed this digest as watermark into the image. If the image is not tampered, the digest at the receiving side must match the one embedded into the image. There are also few schemes that can localize the tampers. They have the advantage of detecting the location of changes; therefore if the changes are located in an unimportant part of the image the need for retransmission is eliminated. The rest of the paper is organized as follows:
This scheme works in two phases: the text processing phase and the image processing phase as shown in figure 1. In text processing phase stream of EPR encoded bits along with hash code bits are prepared and in image processing phase these bits are embedded into the RONI region of corresponding medical image. A. Preparing payload In this technique firstly Class Dependent Coding Scheme (CDCS) is applied on EPR data bits to reduce the size of EPR data bits and enhance the embedding capacity. The CDCS technique assigns fixed codes to each character according to their occurrence probability [12, 13]. In this scheme EPR characters are categorized into three different non-overlapping classes. Class A is considered to be most frequently appearing character set, Class B as an average frequency appearing character set and Class C as a less frequently appearing character set. The number of bits needed to represent each character in the respective class is calculated by assuming only capital letters, alphanumeric and few special characters. Then variable length code is designed to represent each class based on Huffman encoding. Using this scheme any character can be represented by only 4-bits prefixed by a class code (1-bit or 2-bit). CDCS combines the advantage of both fixed length and variable length coding to get less number of bits to represent same information.
In Section II the reversible watermarking technique with high capacity EPR data hiding using CDCS with ROI tamper detection is reviewed. In section III reversible watermarking using two-dimensional difference expansion scheme is discussed. Section IV gives an overview of reversible data hiding using Pixel Difference and Histogram shifting technique. The key features and capabilities of all three techniques is discussed in Section V.
33
International Journal of Emerging Technology and Advanced Engineering Website: www.ijetae.com (ISSN 2250-2459, Volume 2, Issue 1, January 2012) To enhance the robustness of the technique against various attacks such as image compression redundancy bits are added. Tampering along with bit error correction can be achieved by adding redundancy for each bit prior to embedding [12, 13]. Interleaving of bits is done to spread subsequent bits from each other throughout the image, so that even if any block of stego image undergoes with an attack, EPR bits can be recovered from another blocks [11]. These two techniques along with CDCS provide more robustness. The 8x8 blocks of the ROI part of image are not considered for embedding so a secure hash function using SHA-256 is calculated on these ROI blocks [12]. These hash bits are appended to EPR data bits to get the Final Bit Stream (FBS). This FBS is embedded in NROI 8x8 blocks.
D. Retrieval of EPR data To extract the EPR data from stego image firstly it is divided into 8×8 non-overlapping blocks excluding ROI blocks. The difference value α is calculated for each block using above equation [14]. If the calculated difference value lies outside the threshold, then bit 1 is extracted and pixel value of one subset is restored to its original value [14]. If the difference value α lies within threshold, then bit 0 is extracted without changing the pixel value of that block. So following this we can extract EPR data and embedded hash bits. This hash value is then compared with the new hash calculated over stego image‟s ROI to check the authenticity of ROI of image and to check whether it is tampered or not. Original image can be restored without any distortion after EPR data is extracted, after extracting all bits of EPR information it can be reconstructed using CDCS.
B. Finding robust difference parameter ‘α’ We split each 8×8 RONI image block into two subsets A and B. For each of this RONI block, we calculate the difference value α which is defined as the arithmetic average of difference of pixel pairs within the block [14].
III. TWO-DIMENSIONAL DIFFERENCE EXPANSION BASED SCHEME In this scheme 2-Dimensional Difference Expansion (2D-DE) is used which is modified form of previous difference expansion (DE) techniques. This scheme allows embedding the recovery information of ROI without lossy compression which was used previously. This technique can be used for authenticating ROI, hiding patient‟s data, finding tampered areas inside ROI, and recovering those tampered areas. Also, the original image is recovered exactly after watermark extraction at the receiver end [3]. In this scheme, the image is divided into non-overlapping blocks of 4×4 pixels, then using Haar wavelet transform in horizontal and vertical direction (which represent two dimensions of the block) each block is transformed into frequency domain. Then in the blocks which do not cause overflow or underflow 16 bits are embedded using difference expansion [3].
The difference value α is expected to be very close to zero due to correlation and spatial redundancy in the pixel values of local block [14]. We select this α as a robust quantity for embedding information bit as the value „α‟ is based on the statistics of all pixels in the block, it has certain robustness against attacks. C. Embedding bits According to the distribution of α in the original medical image the threshold value„t‟ is chosen. Two cases arise: In case 1, difference value α is located within defined threshold. In this case if 1 is to be embedded, we shift the difference value α to the right side beyond a threshold, this is done by adding or subtracting a fixed number„d‟ called offset from each pixel value within one subset. If 0 is to be embedded, the block is intact [14]. In case 2, if the difference value α is located outside the threshold, no matter whether we have to embed bit 1 or 0, we always embed bit 1, thus shifting the value α farther away beyond the threshold. The bit error introduced in this case is corrected by using error correction [14]. While adding or subtracting„d‟, the pixel values should not lead to overflow/underflow problem.
A. Embedding the payload In this technique first of all ROI region of medical image is selected, RONI and the border regions are defined. ROI is divided into 16×16 pixel blocks. Then MD5 algorithm is used to calculate hash message (H) for ROI. The bits of ROI pixels are collected as P. The LSB‟s of border‟s pixels are collected as L. The patient data, D, is concatenated with P, H and L and is then compressed using Huffman coding to generate the payload. This payload is embedded into RONI using 2-Dimensional Difference Expansion (2D-DE) technique and embedding map is generated.
34
International Journal of Emerging Technology and Advanced Engineering Website: www.ijetae.com (ISSN 2250-2459, Volume 2, Issue 1, January 2012) This embedding map and ROI coordinates are concatenated, compressed and then embedded into LSB‟s of border‟s pixels. The watermarked image is now ready and can be stored or transferred safely [3].
We have used only one peak point for data hiding but for large hiding capacities above process is repeated and every peak point is noted. For communicating multiple peak point binary tree structure is designed.
B. Extraction process
A. Binary tree structure
In the extraction process LSB‟s of the border‟s pixels are recovered. This recovered bit stream is depressed; the embedding map and ROI coordinates are extracted separately. ROI and RONI regions are defined using ROI coordinates. Then the payload is extracted from RONI and is decompressed and then decomposed into; H, P, L, D. Then the hash message of ROI is calculated and compared to extracted one. If they are equal, the image is authentic and then using the bits of L, the LSB‟s of the border‟s pixels are recovered. Else if they are not equal image is unauthentic and it means some tampering is detected. Then to localize the tampered area and to recover the original ROI bits, ROI is divided into blocks of 16×16 pixels. The average value of each block is calculated and matched with value of the average of corresponding pixel in extracted ROI; P. If both are not equal, the block is marked as tampered and is replaced by corresponding pixel P [3]. So in the end, the original image is extracted exactly after watermark extraction.
We assume that we need peak points for embedding a message, where L is level of binary tree and each node is peak point. If pixel difference is less than the peak value message bit can be embedded. If „0‟ bit is added left sub node of binary tree is visited otherwise right sub node is visited. As the payload size is increased tree levels are also increased and recipient needs to share with sender only tree level L [10]. B. Basic embedding process Pixel modification may lead to overflow or underflow, to avoid this we narrow histogram by shifting both sides by units that is histogram is narrowed in the range , 255[10]. This histogram shifting information is recorded as overhead book and embedded into image itself along with payload. In the basic embedding process first we find level L of the binary tree, which determines how much data can be hidden. As stated above histogram is shifted from both sides. To find the difference between adjacent pixel values scan the whole image in an inverse s-order. Again scan the image in inverse s-order to determine if be shifted by . If difference is less than , hide a message bit with [10].
IV. PIXEL DIFFERENCE AND HISTOGRAM SHIFTING BASED SCHEME In the earlier histogram-based reversible data hiding techniques, the message is embedded into histogram bin. In those techniques the pairs of peak and zero points are communicated to the recipients. In this scheme histogram based technique is extended to histogram modification technique using pixel differences. In this technique binary tree is used to eliminate the requirement of communicating pairs of peak and zero point to the recipient and histogram shifting technique is used to prevent overflow and underflow. Neighbour pixels are often highly correlated and have special redundancy; this enables large data capacity while keeping embedding distortion low [10].
C. Process of Extraction In the extraction process at receiver side original image is recovered with the help of L level of a binary tree. Firstly watermarked image is scanned in inverse s-order. Then difference between adjacent pixels is calculated to extract the message bits. The original host image is recovered by shifting in reverse order as done during embedding process. Repeat this process until embedded message is extracted completely. In the end overhead information is extracted which gives us histogram shifting information [10]. So this technique presents an efficient extension of the histogram modification technique by considering the differences between adjacent pixels rather than simple pixel value.
In this scheme we consider N-pixel 8-bit grayscale host image with pixel value denoting the grayscale value of ith pixel. Firstly to calculate differences between adjacent pixels image is scanned in inverse s-order. Then determine the peak point from pixel differences. By scanning the image again in inverse s-order we will check whether the difference is higher than the peak value or not. If difference is higher, message bit cannot be embedded and is shifted by 1 unit, and if it is less a message bit is embedded [10].
V.
DISCUSSION ON FEATURES AND CAPABILITIES
EPR data hiding using Class Dependent Coding Scheme can be used as effective coding scheme for hiding information in medical images which provides better perceptual quality of stego image along with increase in embedding capacity. 35
International Journal of Emerging Technology and Advanced Engineering Website: www.ijetae.com (ISSN 2250-2459, Volume 2, Issue 1, January 2012) This technique enhances the robustness against various attacks like JPEG compression, image tampering and image manipulation [11]. This technique also provides flexibility to doctors in selecting the critical area of the medical image as ROI. Also any tamper in the ROI can be easily detected [11]. If lossless compression is applied on stego image and the stego image is not altered before authentication, then hidden data, original image as well as hash bits can be extracted correctly. Else if compression is not so severe then hidden data can be extracted correctly but original image cannot be recovered [11]. So this technique although provide high data hiding capacity but robustness against intentional attacks is reasonable. Reversible data hiding using 2-dimenstional difference expansion technique can be used for hiding patient‟s data in large volume and provide authenticity to medical images. This technique not only detects the locations of tampered areas inside ROI of watermarked medical images but also recover the contents of tampered areas. Very good performance is achieved in the terms of hiding capacity and visual quality using this technique [3]. Reversible watermarking using pixel difference and histogram shifting introduced the concept of binary tree which predetermines the multiple peak points used to embed messages thus, the only information sender and recipient must share is tree level L, eliminating the need for side communication channel for transferring pairs of peak and minimum points between sender and receiver. As this technique is based on pixel difference, it enables to achieve large hiding capacity while keeping embedding distortion low due to high correlation and spatial redundancy between neighbour pixels [10].
Wiejian Xu, Hui Huang, Wenxue Yu,“A Study of Medical Image Tamper Detection” International Conference of Medical Image Analysis and Clinical Applications, IEEE Transactions on, 2010.
[2]
R.Velumani, V.Seenivasagam,“ A Reversible Blind Medical Image Watermarking Scheme for Patient Identification, Improved Telediagnosis and Tamper Detection with a Facial Image Watermark ” IEEE Transactions on,2010.
[3]
Osamah M. Al-Qershi, Bee Ee Khoo,“ ROI-Based Tamper Detection and Recovery for Medical Images Using Reversible Watermarking Technique” IEEE transaction on, 2010.
[4]
D. Anand and U. C. Niranjan, "Watermarking medical images with patient information," in proceedings of the 20th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 1998, pp. 703-706.
O. M. Al-Qershi and B. E. Khoo, "Authentication and Data Hiding Using a Reversible ROI-based Watermarking Scheme for DICOM Images," in Proceedings of International Conference on Medical Systems Engineering (ICMSE), 2009,pp. 829-834.
[6]
O. M. Al-Qershi and B. E. Khoo, "Authentication and Data Hiding Using a Hybrid ROI-Based Watermarking Scheme for DICOM Images " Journal of Digital Imaging, (Online first) 2009.
[7]
Jasni M Zain, Abdul R M Fauzi, Azian A Aziz, Clinical Evaluation of Watermarked Medical Images, Proceedings of the 28 th IEEE EMBS , Annual International Conference,pp 5459-5462; 2006.
[8]
Nagaraj V.Dharwadkar & B.B Amberker, Supriya & Prateeksha B. Panchannavar,“ Reversible Fragile Medical Image Watermarking with Zero Distortion” International Conference on computer science and communication technology, IEEE Transaction 2010.
[9] Nova Hadi lestriandoko,Taufiq Wirahman,“Reversible Watermarking Using Difference of Virtual Border for Digital image Protection” International conference on Distributed Frameworks for Multimedia Applications(DFmA), IEEE transaction on,2010. [10]
S.L.V.Krishna,B.Abdul Rahim,Fahimuddin Shaik and K.Soundara Rajan,“ Lossless Embedding Using Pixel Differences and Histogram Shifting Technique” IEEE transaction on,2010.
[11]
Sunita V.Dhavale, L.M.Patnaik,“ High Capacity, Robust Lossless EPR Data Hiding Using CDCS with ROI Tamper Detection” Int‟l Conf. on computer and communication Technology,IEEE Transaction on, 2010.
[12]
Osamah M. Al-Qershi, Khoo Bee Ee, “Authentication and Data Hiding Using a Reversible ROI-based Watermarking Scheme for DICOM Images”, World Academy of Science, Engineering and Technology 50 2009.
[13] Suresh N. Mali, Sunita V. Dhavale, “High Capacity Secured Adaptive EPR Data Hiding with Integrity Checking using CDCS”, International Journal of computational sciences, vol.3 no6.December 2009, pp, 657- 668.
REFERENCES [1]
[5]
[14]
36
Zhicheng Ni, Yun Q. Shi, Fellow, Nirwan Ansari, Wei Su, “Robust Lossless Image Data Hiding Designed for Semi-Fragile ImageAuthentication”, IEEE Transactions on circuits and systems for video technology, Vol. 18, No. 4, April 2008, pp. 497-509.