Page 52 - 2025S

UEC Int’l Mini-Conference No.54 45

Algorithm 1 Pixel-to-Block (PTB) Transformation Algorithm 2 Checksum Computation
Require: Original image I of size m×n pixels Require: 2×2 block B
Ensure: Expanded image I exp of size 2m × 2n pix- Ensure: 2-bit checksum C
els 1: sum ← 0
1: for i ← 1 to m do 2: for each pixel p ∈ B except seed do
2: for j ← 1 to n do 3: sum ← sum+(p ≫ 2) ▷ Discard 2 LSBs
3: p ← I(i, j) ▷ Get original pixel value 4: end for
4: I exp (2i−1,2 j −1) ← p ▷ Top-left (seed 5: C[0] ← (sum ≫ 1)&1
pixel) 6: C[1] ← sum&1
5: I exp (2i−1,2 j) ← p+δ 1 ▷ Top-right 7: return C
6: I exp (2i,2 j −1) ← p+δ 2 ▷ Bottom-left
7: I exp (2i,2 j) ← p+δ 3 ▷ Bottom-right Although this scheme is fragile—as the water-
8: end for mark may become corrupted under typical image
9: end for processing attacks such as JPEG compression, rota-

10: return I exp tion, Gaussian noise, or histogram equalization—we
demonstrate that robustness can be enhanced by
The embedding process is performed over a pre- increasing the proportion of the image bandwidth
processed cover image obtained by applying a pixel- used for watermark embedding. In scenarios where
to-block (PTB) transformation, which expands each the watermark occupies a significant portion of the
pixel of the original medical image into a 2×2 block, available space and no pseudorandom walk is ap-
increasing the embedding capacity. Within these plied, the extracted anomaly mask clearly highlights
blocks, data bits are embedded into the least sig- regions affected by tampering.
nificant bits (LSBs) of all pixels except the top- Moreover, due to the small size of the watermark
left one, which is reserved to preserve structure. A and the structured embedding strategy, even when
lightweight checksum is also computed and embed- foreign objects are overlaid or certain areas of the
ded in the second LSB of the bottom row to allow image are modified, the watermark remains partially
integrity verification. recoverable. This residual visibility is especially ef-
fective for tampering localization, as it allows us to
Embedding Capacity. The total hiding capacity identify suspicious regions without compromising
of the system is estimated using the dimensions of a the integrity of the medical content.
standard high-resolution DICOM image as follows: Encryption and decryption are synchronized
through the chaotic map without requiring key ex-
change, and the full embedding-extraction process
7040×8560
Capacity = ×15 bits = 56,496,000 bits maintains a high level of image fidelity. This is evi-
4
(4) denced by the quantitative results obtained.
This estimation assumes 15 usable LSBs per 2×2
block, excluding the seed pixel. 3.2.1 Robustness Evaluation

Insertion Summary. The embedding process Table 1 presents the recovery performance of the
comprises the following steps: embedded data under various image modifications.
• PTB transformation: Converts each pixel into
a 2×2 block to increase capacity (Algorithm 1). 3.2.2 Performance and Clinical Evaluation

• Data embedding: Bits are inserted into 15 In terms of computational and clinical feasibility, the
LSBs of each block (excluding top-left seed). proposed method shows strong performance:

• Checksum insertion: A lightweight checksum • Embedding time: 93.00 seconds on Apple M2
is embedded in the second LSB of the bottom chip (compared to 112 seconds for difference-
row for later tamper detection. expansion schemes)