Academia.eduAcademia.edu

Figure from:A High-Capacity 3D Steganography Algorithm With Adjustable Distortion

See full PDFdownloadDownload figure
FIGURE 9. Cases the proposed method failed to handle: the models with multi-axial symmetry models, such as (a) sphere and (b) star. ' Note that Niayer Tepresents the number of layers, a is constant in [26], normal degradation tolerances « = 10 in [27], symbols ’X’,’,/’ and ‘A’ indicate that the approach can not withstand, can withstand, and can weakly withstand attacks, respectively.

Figure 9 Cases the proposed method failed to handle: the models with multi-axial symmetry models, such as (a) sphere and (b) star. ' Note that Niayer Tepresents the number of layers, a is constant in [26], normal degradation tolerances « = 10 in [27], symbols ’X’,’,/’ and ‘A’ indicate that the approach can not withstand, can withstand, and can weakly withstand attacks, respectively.

Related Figures (9)

FIGURE 3. An illustration on the distortion of stego-models with different number of embedded data (k = 1, 5, 10, 20, 30). Both the experimental distortion (top) and the corresponding worst distortion (bottom) are shown using the same color bar (the largest distortion is 0.17 x 10-5).
FIGURE 3. An illustration on the distortion of stego-models with different number of embedded data (k = 1, 5, 10, 20, 30). Both the experimental distortion (top) and the corresponding worst distortion (bottom) are shown using the same color bar (the largest distortion is 0.17 x 10-5).
FIGURE 4. The flowchart of the secret extraction procedure.
FIGURE 4. The flowchart of the secret extraction procedure.
TABLE 3. Relationship between the embedding k and the PSNR on various models (/ = 6).
TABLE 3. Relationship between the embedding k and the PSNR on various models (/ = 6).
FIGURE 5. The relationship between / and the performances (PSNR and ER). PSNR grows with the increase of / (a) at the cost of certain ER (in the case of maximum embedding capacity with the corresponding k) (b).
FIGURE 5. The relationship between / and the performances (PSNR and ER). PSNR grows with the increase of / (a) at the cost of certain ER (in the case of maximum embedding capacity with the corresponding k) (b).
Moreover, we could also find that the performances have almost no effects associated with the quality of models due to the utility of the truncated space, take an example, the models with and without noise have almost the same embedding capacity and PSNR. As shown in Fig. 8, the performances are demonstrated in (a) Bunny (PSNR = 133.15, ER = 92.99), (b) Bunny with noise (PSNR = 133.14, ER = 92.99),
Moreover, we could also find that the performances have almost no effects associated with the quality of models due to the utility of the truncated space, take an example, the models with and without noise have almost the same embedding capacity and PSNR. As shown in Fig. 8, the performances are demonstrated in (a) Bunny (PSNR = 133.15, ER = 92.99), (b) Bunny with noise (PSNR = 133.14, ER = 92.99),
FIGURE 7. The comparisons with the steganography schemes: Chao ef al.’s scheme [36], Lin et al.’s scheme [26] and Yang et al.'s [27], in terms of embedding rate and the PSNR for different models (Horse and Gargoyle, / = 6).
FIGURE 7. The comparisons with the steganography schemes: Chao ef al.’s scheme [36], Lin et al.’s scheme [26] and Yang et al.'s [27], in terms of embedding rate and the PSNR for different models (Horse and Gargoyle, / = 6).
FIGURE 8. The embedding capacity of the proposed method is not affected by the quality of models. Both original models and the models with noise (100% mean edge length) are used to test the performances (/ = 6, k = 31).
FIGURE 8. The embedding capacity of the proposed method is not affected by the quality of models. Both original models and the models with noise (100% mean edge length) are used to test the performances (/ = 6, k = 31).
TABLE 4. The comparison between our method (/ = 6, k = 31) and the methods in [26], [27], [34], [36], and [37].
TABLE 4. The comparison between our method (/ = 6, k = 31) and the methods in [26], [27], [34], [36], and [37].