Page 56 - 2024-Vol20-Issue2
P. 56
52 | Alobaidi & Mikhael
(a) (b) (a) (b)
(c) (d) (c) (d)
(e) (f) (e) (f)
Fig. 2. Sample 1 image from custom-built database/proposed Fig. 3. Sample 2 image from custom-built database/proposed
LSB technique. (a) Original cover image, (b) Stegoimage LSB technique. (a) Original cover image, (b) Stegoimage
with message size of 6KB, (c) Stegoimage with message size with message size of 6KB, (c) Stegoimage with message size
of 8KB, (d) Stegoimage with message size of 10KB, (e) of 8KB, (d) Stegoimage with message size of 10KB, (e)
Stegoimage with message size of 14KB, (f) Stegoimage with Stegoimage with message size of 14KB, (f) Stegoimage with
message size of 16KB. message size of 16KB.
black and white images for experiments with detecting To examine the effect of the block size, or dimensions, on the
steganographically hidden data in JPEG images. It con- performance of the proposed technique, different block sizes
tains discrete cosine transform residuals (DCTR), Ga- are considered. As shown in Tables I through XII, the block
bor filter residuals (GFR) and PHase Aware pRojection size of 4 means 4 × 4 which is 16 pixels, or coefficients. First,
Model (PHARM) features extracted out of clean im- 1 sample from the database is shown in Fig. 6 that shows
ages, images with random data hidden using the JPEG the original image besides the stegoimages for different block
Universal Wavelet Relative Distortion, images with ran- sizes for the LSB Spatial case. Secondly, DCT block insertion
dom data hidden using the nsF5 method, images with case outputs are shown in Fig. 7. Finally, the proposed
random data hidden using the Uniform Embedding Re- technique outputs are shown in Fig. 8.
visited Distortion (UERD) algorithm.
The results (in terms of PSNRs)for this category are V. DISCUSSION
shown in Tables I through VI. On the other hand, Tables
VII through XII are populated with RMSE. The SSIM A. BOSSBase Results
for most of the results were close to unity. As shown in the presented results, the proposed technique per-
formed better than the other two techniques under comparison.
2. Custom-built image database appeared in [12] The RMSE maintained at lower levels while higher PSNRs
Four grey-scale image, shown in Figs. 2 through 5, are achieved. The visual human inspection illustrate that the
form the second tested database. The results (in terms proposed technique does not alter the visual properties of the
of PSNRs)for this category are shown in Table XIII. cover image. In terms of utilized performance metrics, the
On the other hand, Table XIV is populated with RMSE. best block size (window dimensions) is 4 × 4 (16 coefficients).
The SSIM for most of the results were close to unity.
