Multiple Image Watermarking Applied to Health Information Management Reporter :黃阡廷 1 Outline Introduction Proposed Method Algorithm Selection of Embeddable Coefficients Results BCH encoding PSNR & wPSNR NHD Conclusions 2 introduction IEEE Transactions on Information Technology in Biomedicine, VOL. 10, NO. 4, October 2006 Author: Aggeliki Giakoumaki, Sotiris Pavlopoulos, Dimitris Koutsouri Research Motivation and Background: -huge and exponentially increasing amount of medical data -sensitive nature of patients’ personal data Research Purpose: -potentials of digital watermarking in medical data management issues -proposes a multiple watermarking scheme regarding health data handling 3 introduction Research Method: -Haar wavelet transform -quantization function -multiple watermarks embedding procedure data watermarks: signature, index, caption watermark -energy of approximation -BCH encoding schemes S. Zinger, Z. Jin, H. Maitre, and B. Sankur, “Optimization of Watermarking performances using error correcting codes and repetition” -peak signal-to-noise ratio (PSNR) -weighted PSNR (wPSNR) -noise visibility function (NVF) -normalized hamming distance (NHD) 4 introduction regions of interest (ROI) Medical Data Management Issues: -Access Control -De-identification -Captioning -Origin Identification -Integrity Control -Indexing 5 Proposed Method dyadic scaling decomposition of the wavelet transform and the signal processing of the human visual system (HVS) signature watermark: source authentication by the recipient index watermark: image retrieval by database querying mechanisms caption watermark: additional data useful for the diagnosis reference watermark: data integrity control and tampering localization 6 Algorithm Haar wavelet transform produces coefficients that are dyadic rational numbers: 2l quantization function: -k is an integer -s is a user-defined offset for increased security -Δ, the quantization parameter, is a positive real number 7 Algorithm quantization parameter Δ is defined as: Δ = 2l , where l is the decomposition level. 8 Algorithm The multiple watermarks embedding procedure: Step 1: four-level Haar wavelet decomposition Step 2: watermark bit wi is embedded into the coefficient f according to the following: a) If Q (f ) = wi , the coefficient is not modified. b) following assignment: Q (f ) =wi Step 3: The watermarked image is produced by the corresponding four-level inverse wavelet transform. 9 10 Selection of Embeddable Coefficients signature watermark is embedded in the fourth decomposition level index watermark is embedded in the third decomposition level caption watermark is embedded in the second decomposition level the first decomposition level is used for fragile watermarking to allow data integrity control reference watermark is embedded in selected coefficients of the other three decomposition levels 11 Energy of Approximation and Detail Images of a Four-Level Wavelet Decomposition k denotes the approximation and the detail images at each of the decomposition levels Ik are the coefficients of the subband images Nk and Mk are their corresponding dimensions 12 Allocation of Watermarks According to Robustness and Capacity Criteria 13 Results test set consisted of 50 ultrasound images of size 256×320 pixels signature watermark containing the doctor’s identification key is a 128-b watermark reference watermark is a binary array index and caption watermarks are binary arrays produced by the ASCII codes of text files set of keywords consisted of six words and a total of 52 characters patient’s data comprised of 23 words, of 208 characters in total 14 BCH encoding In order to increase robustness of the embedded data (signature, index, caption), error correction coding was implemented. BCH encoding schemes: BCH(n, k, l ) n : codeword of length k : bits of the watermark array l : can correct bit errors 15 PSNR & wPSNR (a) Original image. (b) Resulting watermarked image. 16 PSNR & wPSNR I : original image I hat : watermarked image N I : the number of pixels in the image maxI (m, n): the maximum gray value of the original image 17 PSNR & wPSNR The weighted PSNR (wPSNR) is a quality metric that assigns different weights to the perceptually different image regions, based on the noise visibility function (NVF). For flat regions, the NVF value is close to 1, whereas for edge or textured regions, it is closer to 0. 18 PSNR & wPSNR 19 NHD Normalized Hamming Distance (NHD) w : the original watermark w hat : extracted fragile watermark Nw : the length of the watermark 20 Percentage of Error Bits in Extracted Watermarks 21 (a) Ultrasound image with a blurred region. (b) Tampering detection through the difference image of the 1st decomposition level reference watermark. method has been tested on other medical imaging modalities namely MRA, CT, MRI, and PET, and the results were also satisfactory 22 Conclusions Digital watermarking has the potential to provide complementary and alternative solutions in a range of issues of critical importance to health informatics. The experimental results demonstrate the efficiency of the scheme, which could be extended and integrated into healthcare information systems. Future work involves integration of the watermarking scheme with JPEG2000 compression. 23