Page 48 - 2025S

UEC Int’l Mini-Conference No.54 41

preservation of edge and texture information [8]. marking in medical images, contributing to both
Then, instead of using artificial vectorized contours data security and patient authentication.
to increase embedding bandwidth, we introduce a
segmentation-based framework that isolates the sub- 2 Background
ject and surrounds it with a programmatically gener-
ated background of vertical white lines—optimized 2.1 Halftone technique
for binary data insertion.
Furthermore, we adopt the fragile-watermarking Halftoning is a technique that converts grayscale im-
scheme for secure embedding into medical images. ages into binary representations that are visually per-
The halftoned patient photo is encrypted using a lo- ceived by the human eye as continuous tones. Un-
gistic chaotic map and embedded into the 12th to like traditional binarization methods, which rely on
16th bit-planes of DICOM images using a pseudo- a fixed threshold to determine pixel values, halfton-
random walk mechanism, enhancing security and re- ing distributes black and white dots in such a way
sistance to unauthorized extraction. The watermark that varying densities simulate different gray levels.
integrity is verifiable using a lightweight checksum There are two main categories of halfton-
computed for each embedded block, allowing for ing: amplitude-modulated (AM) and frequency-
tamper localization in the case of data corruption. modulated (FM). In AM halftoning, the size of each
To further improve the quality of reconstructed dot varies according to the local intensity of the
images, particularly in the halftone-to-grayscale re- grayscale value, while the spatial frequency remains
covery process, we integrate a DnCNN (Denoising fixed. In contrast, FM halftoning uses uniformly
Convolutional Neural Network) to reduce binariza- sized dots but varies their spatial distribution de-
tion noise. This enhances the authenticity of patient pending on intensity. This study focuses on FM
photographs extracted from DICOM images. halftoning due to its superior ability to reproduce
The main contributions of this proposal are sum- fine details and gray variations through precise spa-
marized as follows: tial control.
Halftoning has been widely adopted in fields such
1. A reversible data hiding scheme that links a pa- as steganography, cryptography, and media com-
tient’s RGB image to their DICOM file by con- pression, adapting to the needs of modern display
verting the photo into a Jarvis halftone image, and printing technologies. While AM techniques
embedding it into higher LSB layers (bits 12- were historically favored for producing smooth tonal
16) using a pseudo-random walk and chaotic gradients, FM halftoning has become the preferred
encryption. method in industrial applications due to its enhanced
visual accuracy and resolution [9].
2. A halftone embedding capacity boost through
segmentation-based background enhancement: 2.2 Error Diffusion Halftoning Method
an artificial border composed of white vertical
lines is added to the patient’s silhouette, allow- The error diffusion method is a widely used halfton-
ing binary clinical information to be concealed ing technique that binarizes grayscale images by ap-
using a high-payload data hiding algorithm for plying a quantization function, denoted as Q, based
binary images. on a predefined threshold. Each pixel is compared
against this threshold to determine whether it should
3. A two-step image reconstruction strategy that be rendered as black or white. This process is il-
restores the halftone image to grayscale. This lustrated in Figure 1 and mathematically defined by
involves downscaling and then enhancing the Equation (1).
image using DnCNN-based super-resolution
techniques, recovering visually accurate patient ( 0 if u(i, j) < Th,
images for identification. b(i, j) = Q(u(i, j)) = (1)
1 if u(i, j) ≥ Th.
In doing so, the proposed method ensures high- Once a pixel is quantized, the quantization
capacity, reversible, and tamper-detectable water- error—i.e., the difference between the original