Academia.eduAcademia.edu

Figure 3 – uploaded by monalisa sahu

See full PDFdownloadDownload figure
Figure 3: General structure of the cryptographic communication process.

Figure 3 General structure of the cryptographic communication process.

Related Figures (41)

Table 1: Comparision of the working principles for cryptography, watermarking, and Steganography
Table 1: Comparision of the working principles for cryptography, watermarking, and Steganography
Figure 1: Classification of data security techniques
Figure 1: Classification of data security techniques
Figure 2: Displaying year-wise publications considered in this review. during World War II. The invention of the microdot is cred- ited o the Germans, and it was they who were using it with balloons and pigeons. In the microdot technique, the se- cret messages can be reduced to a very small dot (close to 1 mm and embedded in another message. Later the dot can be enlarged to retrieve the original message. One such fas- cinating idea was to send the messages using ideograms or id eographs. An ideogram is just a symbol to represent a given message. In the ancient days, it was a popular em. bedding technique in China. History also tells, steganogra- phy was used in many forms, such as inserting a message in a postage stamp, null cipher, and morse code, etc. These are a few notable steganography techniques that people were using mostly before digitization. and future scopes of IS. Finally, the concluding remarks are drawn in Section 7.
Figure 2: Displaying year-wise publications considered in this review. during World War II. The invention of the microdot is cred- ited o the Germans, and it was they who were using it with balloons and pigeons. In the microdot technique, the se- cret messages can be reduced to a very small dot (close to 1 mm and embedded in another message. Later the dot can be enlarged to retrieve the original message. One such fas- cinating idea was to send the messages using ideograms or id eographs. An ideogram is just a symbol to represent a given message. In the ancient days, it was a popular em. bedding technique in China. History also tells, steganogra- phy was used in many forms, such as inserting a message in a postage stamp, null cipher, and morse code, etc. These are a few notable steganography techniques that people were using mostly before digitization. and future scopes of IS. Finally, the concluding remarks are drawn in Section 7.
Figure 4: Applications of cryptography.
Figure 4: Applications of cryptography.
Figure 7: Applications of digital watermarking. Figure 6: Watermarking extraction process. other hand, the robust wa ex some a ues are commonly app niq na fits from both robust and ract ly, t niq cat ion. ues are primarily used f Further, the semi-fragile watermark techniques are teration to the wa he semi-fragile wa he watermark accurately even after there has been ermarking technique is able to ermark. Therefore, these tech- ied for copyright protection. Fi- ermark technique takes bene- fragile watermarks. These tech- or both integrity and authenti- able to detect the tempered area localization and recovery.
Figure 7: Applications of digital watermarking. Figure 6: Watermarking extraction process. other hand, the robust wa ex some a ues are commonly app niq na fits from both robust and ract ly, t niq cat ion. ues are primarily used f Further, the semi-fragile watermark techniques are teration to the wa he semi-fragile wa he watermark accurately even after there has been ermarking technique is able to ermark. Therefore, these tech- ied for copyright protection. Fi- ermark technique takes bene- fragile watermarks. These tech- or both integrity and authenti- able to detect the tempered area localization and recovery.
Figure 7 shows the various applications of watermarking in real life. 2.3 Digital steganography
Figure 7 shows the various applications of watermarking in real life. 2.3 Digital steganography
Figure 9: Applications of steganography. Figure 8: General structure of the steganographic communication process. The steganographic process uses various digital ob- jects as the carrier signals such as image, text, DNA, video, audio, and text, etc. Due to the property of innocence of digital images, researchers have preferred images as the carrier signal for hiding secret information. Also, the pres- ence of redundant pixels in an image makes it even more suitable for embedding secret information. Hiding confi- dential information inside the image is known as image steganography (IS). Figure 10 shows various domains in which steganography can be operated. However, most of the techniques presented in the literature were highly fo- cused on either the spatial or transform domain.
Figure 9: Applications of steganography. Figure 8: General structure of the steganographic communication process. The steganographic process uses various digital ob- jects as the carrier signals such as image, text, DNA, video, audio, and text, etc. Due to the property of innocence of digital images, researchers have preferred images as the carrier signal for hiding secret information. Also, the pres- ence of redundant pixels in an image makes it even more suitable for embedding secret information. Hiding confi- dential information inside the image is known as image steganography (IS). Figure 10 shows various domains in which steganography can be operated. However, most of the techniques presented in the literature were highly fo- cused on either the spatial or transform domain.
and less time-consuming. On the other hand, transform domain techniques utilize the frequency content, and they are based on orthogonal transformation (frequency and phase) to the image. In the transform domain, apply- ing various transformations and inverse transformations, such as Fourier, Laplace, and Z the em carried out. Some common transform d are (1) discrete Fourier transformation bedding process is omain techniques (DFT) (2) discrete wavelet transformation (DWT) (3) discrete cosine transfor- mation (DCT), and (4) singular value d ecomposition. Fig- ure 11 displays the classification of spa ial domain ISTs.
and less time-consuming. On the other hand, transform domain techniques utilize the frequency content, and they are based on orthogonal transformation (frequency and phase) to the image. In the transform domain, apply- ing various transformations and inverse transformations, such as Fourier, Laplace, and Z the em carried out. Some common transform d are (1) discrete Fourier transformation bedding process is omain techniques (DFT) (2) discrete wavelet transformation (DWT) (3) discrete cosine transfor- mation (DCT), and (4) singular value d ecomposition. Fig- ure 11 displays the classification of spa ial domain ISTs.
Table 2: Requirements for an efficient IST. an efficient and competent IST are mentioned in Table 2. Following this, the performance appraisal metrics for the ISTs are presented below in Table 3.
Table 2: Requirements for an efficient IST. an efficient and competent IST are mentioned in Table 2. Following this, the performance appraisal metrics for the ISTs are presented below in Table 3.
The primary objectives for any ISTs are to simultaneously ihe primary objec achieve (1) high HC, (2) better imperceptibility, and (3) grea Ives for any lols are to simultaneously er robustness or security. The HC refers to the abil- ity to hide the maximum number of secret data. It is usu- ally ity. In o OI. 1 computed in The CI and O her words, th bits. Im should subtle be visually indis e CI should be impercepti The quality of using various image quality assessment me (1) mean square error (MSE), (2) peak signa tio ( PSNR), (3) weighted PSNR (WPSNR) (4 perceptibility suggests CI qual- inguishable. ble from the y in the CI can be computed rics such as -to-noise ra- ) root mean square error (RMSE), (5) universal image quality index (Q), 6) structural similarity index (SSIM), (7 normalized cross-correlation (NCC), (8) Kullback-Leibler (KL) diver- gence (Dx,;), (9) Manhattan distance (MD) and (10) Eu- clidean distance (ED). nm
The primary objectives for any ISTs are to simultaneously ihe primary objec achieve (1) high HC, (2) better imperceptibility, and (3) grea Ives for any lols are to simultaneously er robustness or security. The HC refers to the abil- ity to hide the maximum number of secret data. It is usu- ally ity. In o OI. 1 computed in The CI and O her words, th bits. Im should subtle be visually indis e CI should be impercepti The quality of using various image quality assessment me (1) mean square error (MSE), (2) peak signa tio ( PSNR), (3) weighted PSNR (WPSNR) (4 perceptibility suggests CI qual- inguishable. ble from the y in the CI can be computed rics such as -to-noise ra- ) root mean square error (RMSE), (5) universal image quality index (Q), 6) structural similarity index (SSIM), (7 normalized cross-correlation (NCC), (8) Kullback-Leibler (KL) diver- gence (Dx,;), (9) Manhattan distance (MD) and (10) Eu- clidean distance (ED). nm
Where Foc i is the co-variance, 0 and € are the mean, 04 and o2 are the standard deviation for the corresponding OI and CI. 6, C, 03, 02, and doc are calculated using Equations (15), (16), (17), (18) and (19). Q and SSIM both measure the similarity between OI and CI. The optimum value for both Q and SSIM is 1. Q can be found using Equation (14).
Where Foc i is the co-variance, 0 and € are the mean, 04 and o2 are the standard deviation for the corresponding OI and CI. 6, C, 03, 02, and doc are calculated using Equations (15), (16), (17), (18) and (19). Q and SSIM both measure the similarity between OI and CI. The optimum value for both Q and SSIM is 1. Q can be found using Equation (14).
Table 4: Silent features of various state-of-art review papers that are reported till date.
Table 4: Silent features of various state-of-art review papers that are reported till date.
Table 4: ... continued
Table 4: ... continued
Figure 14a shows the RS plot of the Lena image when there is no data embedded inside the image. This can be clearly observed in the case Rm ~ R-m > Sm ~ Sm. On the contrary, the presence of t clearly noticeable for the sim shown in Figure 14b. Here, Rm — Sm holds well. Similarly, for the CI of the three LSBS he secret information is quite ple 1LSBS technique. This is he condition R-m - S-m > when we plot the histogram echniques with the original one, it is found that the his 1LSBS technique is identical. ogram of OI and CI for the However, the zig-zag nature of the CI can be easily spotted for both 2LSBS and 3LSBS techniques. Figure 15a,b,c shows the respective PDH-plots for the 1LSBS, 2LSBS and 3LSBS techniques.
Figure 14a shows the RS plot of the Lena image when there is no data embedded inside the image. This can be clearly observed in the case Rm ~ R-m > Sm ~ Sm. On the contrary, the presence of t clearly noticeable for the sim shown in Figure 14b. Here, Rm — Sm holds well. Similarly, for the CI of the three LSBS he secret information is quite ple 1LSBS technique. This is he condition R-m - S-m > when we plot the histogram echniques with the original one, it is found that the his 1LSBS technique is identical. ogram of OI and CI for the However, the zig-zag nature of the CI can be easily spotted for both 2LSBS and 3LSBS techniques. Figure 15a,b,c shows the respective PDH-plots for the 1LSBS, 2LSBS and 3LSBS techniques.
Figure 14: RS plots for the Lena image (a) with no embedding information and (b) for 1LSBS technique.
Figure 14: RS plots for the Lena image (a) with no embedding information and (b) for 1LSBS technique.
Figure 15: PDH-plots for Lena OI and Cl for (a) 1LSBS, (b) 2LSBS and (c) 3LSBS.
Figure 15: PDH-plots for Lena OI and Cl for (a) 1LSBS, (b) 2LSBS and (c) 3LSBS.
Figure 16: Respective bit planes (starting from LSB) for the Lena OI (with no embedding information).
Figure 16: Respective bit planes (starting from LSB) for the Lena OI (with no embedding information).
Figure 17: Displaying the produced images from the 1° bit planes of (a) original Lena image (b) 1LSBS (c) LSBM [77], and (d) Wu and Tsai [94].
Figure 17: Displaying the produced images from the 1° bit planes of (a) original Lena image (b) 1LSBS (c) LSBM [77], and (d) Wu and Tsai [94].
Figure 18: An illustration of Wu and Hwang’s [62] technique.
Figure 18: An illustration of Wu and Hwang’s [62] technique.
Table 5: Analysis of LSBS, LSBM, and LSBA based techniques.
Table 5: Analysis of LSBS, LSBM, and LSBA based techniques.
Figure 19: An illustration for Mielikainen’s [77] technique.
Figure 19: An illustration for Mielikainen’s [77] technique.
Figure 20: An illustration of Wu and Tsai’s [94] PVDS technique
Figure 20: An illustration of Wu and Tsai’s [94] PVDS technique
Figure 21: (a) RS-plot and (b) PDH-plot of Lena image for Wu and Tsai’s PVDS [94] technique.
Figure 21: (a) RS-plot and (b) PDH-plot of Lena image for Wu and Tsai’s PVDS [94] technique.
Figure 22: An example of OUP in PVDS [94] embedding process.
Figure 22: An example of OUP in PVDS [94] embedding process.
Table 6: No. of pixels that suffer from OUP for various PVDS based techniques.
Table 6: No. of pixels that suffer from OUP for various PVDS based techniques.
Figure 23: 512x512 grayscale images.
Figure 23: 512x512 grayscale images.
Figure 24: A pictorial illustration of embedding and extraction for Wu et al.’s [117] technique. ep 1: Assume the two CPs ofa block are 0; and 02, and the secret bits are bjb2b3bybsbe¢.
Figure 24: A pictorial illustration of embedding and extraction for Wu et al.’s [117] technique. ep 1: Assume the two CPs ofa block are 0; and 02, and the secret bits are bjb2b3bybsbe¢.
Figure 25: (a) RS-plot and (b) PDH-plot for Wu et al.’s [117] technique for the Lena image
Figure 25: (a) RS-plot and (b) PDH-plot for Wu et al.’s [117] technique for the Lena image
Table 9: Range table of Jung’s [125] technique. 4.3.1 Jung’s LSBS+PVDS technique [125] In 2018, Jung [125] proposed a capacity promotive tech- nique by taking advantage of both LSBS and PVDS strate- gies. The steps of both the embedding and the extraction processes are discussed below. The extraction from the block (cj, c)) is performed us- ng the following steps.
Table 9: Range table of Jung’s [125] technique. 4.3.1 Jung’s LSBS+PVDS technique [125] In 2018, Jung [125] proposed a capacity promotive tech- nique by taking advantage of both LSBS and PVDS strate- gies. The steps of both the embedding and the extraction processes are discussed below. The extraction from the block (cj, c)) is performed us- ng the following steps.
Table 10: Analysis of LSBS + PVDS based techniques.
Table 10: Analysis of LSBS + PVDS based techniques.
Figure 26: An illustration of OUP in Jung’s [125] technique.
Figure 26: An illustration of OUP in Jung’s [125] technique.
Table 11: Analysis of LSBS+PVDS+MF based techniques. 4.4 LSBS+PVDS+MF based steganography nique to PDH analysis. They found abnormal fluctuation in the PVD histogram for the CI with the Wang et al.’s [134] technique. Further, they addressed this issue by propos- ing a novel PVD+MF based technique using turnover pol- icy and pixel readjustment process. Xu et al. [136] sug- gested a novel LSB substitution and modulo three strat- egy technique using the concept of addition and subtrac- tion to embed two ternary numbers in the respective pix- els. This technique outperforms conventional LSB substi- tution techniques. Aimed at optimized CI, Li and He [137] devised a particle swarm optimization based PVDS+MF technique. This technique provides a balanced impercep- Wang et al. [134] were the first who introduced the con- cept of embedding secret bits by utilizing both PVDS and MF strategy. Here, instead of directly modifying the differ- ence between the two pixels, as was done in the conven- tional PVDS [94] technique, they modified the remainder of the two CPs. Further, Wang et al.’s [134] technique uti- lized an optimization strategy to avoid the OUP. The results of the investigation suggest that Wang et al.’s [134] tech- nique offers imperceptible CI and acceptable BPP. Further, this technique showed good ARA to RS analysis. Joo et al. [135] pointed out the weakness of Wang et al.’s [134] tech-
Table 11: Analysis of LSBS+PVDS+MF based techniques. 4.4 LSBS+PVDS+MF based steganography nique to PDH analysis. They found abnormal fluctuation in the PVD histogram for the CI with the Wang et al.’s [134] technique. Further, they addressed this issue by propos- ing a novel PVD+MF based technique using turnover pol- icy and pixel readjustment process. Xu et al. [136] sug- gested a novel LSB substitution and modulo three strat- egy technique using the concept of addition and subtrac- tion to embed two ternary numbers in the respective pix- els. This technique outperforms conventional LSB substi- tution techniques. Aimed at optimized CI, Li and He [137] devised a particle swarm optimization based PVDS+MF technique. This technique provides a balanced impercep- Wang et al. [134] were the first who introduced the con- cept of embedding secret bits by utilizing both PVDS and MF strategy. Here, instead of directly modifying the differ- ence between the two pixels, as was done in the conven- tional PVDS [94] technique, they modified the remainder of the two CPs. Further, Wang et al.’s [134] technique uti- lized an optimization strategy to avoid the OUP. The results of the investigation suggest that Wang et al.’s [134] tech- nique offers imperceptible CI and acceptable BPP. Further, this technique showed good ARA to RS analysis. Joo et al. [135] pointed out the weakness of Wang et al.’s [134] tech-
Table 12: Analysis of edge based techniques. Edge-based steganography utilizes the concept of identi- fying the edge regions and performing the embedding in those regions. The human visual system can tolerate a significant change to the edge or texture regions. There- fore, embedding the secret bits in the edge regions is more productive. In the literature, authors have suggested vari- ous edge detection techniques such as Sobel, Fuzzy, Pre- witt, Canny, Laplacian, and Hybrid [157, 178] to locate the edge regions effectively. However, an optimal selection of edge also depends on some other parameters such as Gaus- sian kernel, gradient descent, and the threshold selection. Sun [157] proposed an improved Canny edge detector (CED) based technique that can observe more edge pixels in an image. Further, using Huffman coding, the HC has been improved significantly. Also, a sorting table has been im- plemented to randomize the edge pixels to increase the ARA. Recently, an improved Fuzzy edge detection (FED) based technique using LSBM to improve the CI quality has been suggested by Dadgostar and Afsari [160]. To improve the HC, the LSBM based technique that can effectively dis- ibility and HC. Later Swain [138, 139] illustrated the range mismatch issue that occurs in Wang et al.’s [134] tech- nique. Also, to address this issue, a novel PVDS+MF+LSBS 138] technique has been suggested. Further, experimen- ally it has been proved that the proposed technique is secure against both RS and PDH analysis. Further, the OUP has been successfully addressed. A new block-based echnique combining MF+PVDS has been proposed by Dhamari and Darabkh [142] to improve the HC and im- perceptibility. Sahu and Swain [149] proposed an optimal data hiding technique using PVDS and MF. There are two variants proposed. Both variants utilize the fluctuation be- ‘ween two CPs for embedding the secret bits. Later, using he MF the pixels are modified to improve the quality of the CI. Results show that variant 1 offers good imperceptibility, and variant 2 offers larger HC. Table 11 presents the results of the average PSNR, BPP, ARA to RS and PDH analysis, and OUP status for various LSBS+PVDS+MF based tech- niques.
Table 12: Analysis of edge based techniques. Edge-based steganography utilizes the concept of identi- fying the edge regions and performing the embedding in those regions. The human visual system can tolerate a significant change to the edge or texture regions. There- fore, embedding the secret bits in the edge regions is more productive. In the literature, authors have suggested vari- ous edge detection techniques such as Sobel, Fuzzy, Pre- witt, Canny, Laplacian, and Hybrid [157, 178] to locate the edge regions effectively. However, an optimal selection of edge also depends on some other parameters such as Gaus- sian kernel, gradient descent, and the threshold selection. Sun [157] proposed an improved Canny edge detector (CED) based technique that can observe more edge pixels in an image. Further, using Huffman coding, the HC has been improved significantly. Also, a sorting table has been im- plemented to randomize the edge pixels to increase the ARA. Recently, an improved Fuzzy edge detection (FED) based technique using LSBM to improve the CI quality has been suggested by Dadgostar and Afsari [160]. To improve the HC, the LSBM based technique that can effectively dis- ibility and HC. Later Swain [138, 139] illustrated the range mismatch issue that occurs in Wang et al.’s [134] tech- nique. Also, to address this issue, a novel PVDS+MF+LSBS 138] technique has been suggested. Further, experimen- ally it has been proved that the proposed technique is secure against both RS and PDH analysis. Further, the OUP has been successfully addressed. A new block-based echnique combining MF+PVDS has been proposed by Dhamari and Darabkh [142] to improve the HC and im- perceptibility. Sahu and Swain [149] proposed an optimal data hiding technique using PVDS and MF. There are two variants proposed. Both variants utilize the fluctuation be- ‘ween two CPs for embedding the secret bits. Later, using he MF the pixels are modified to improve the quality of the CI. Results show that variant 1 offers good imperceptibility, and variant 2 offers larger HC. Table 11 presents the results of the average PSNR, BPP, ARA to RS and PDH analysis, and OUP status for various LSBS+PVDS+MF based tech- niques.
Table 13: ... continued
Table 13: ... continued
Table 14: Analysis of other steganography techniques.
Table 14: Analysis of other steganography techniques.
Table 14: ... continued 5 Steganography and steganalysis towards machine learning network can hide an image inside the OI to produce the CI, and the decoder network can produce the complete hid- den image from the CI. However, the robustness of the tech- nique has not been verified. In [315], a two-stage authen- tication based deep steganography and matrix encoding technique has been not directly embedd the distorti ing technig improve id entity au proposed. Here, the secret message is ed in the pixels of the OI. Therefore, on in the Cl is significantly less than other exist- ues. Primarily, the objective of this schema is to hentication. An invisible steganogra- phy technique associating generative adversarial network and mixed loss func ion has been suggested in [316]. Here, a secret image is embedded in the original color image of equal size. However, to obtain imperceptible CI, the se- cret bits are embedded only in the Y channel, which do not carry any color-related features of the image. A U-Net CNN structure-based IST with two different networks such as hidden and extraction networks has been proposed in [317]. Using the hidden network, the sender embeds the se- cret image in an OI, and the receiver extracts the SI from the CI to obtain the SI. The resulting CI does not exhibit any visual cues to the outsider. towards machine learning With the rapid growth in the field of artificial intelligence, many steganographers move towards ML-based steganog- raphy and steganalysis [312]. It is too early to speak abou whether a trained model can be really helpful in regards to hiding and recovering information from the digital objec accurately. However, recently some researchers proved hat it is possible for machines to hide the secret message in a digital image. Until now, the count of such techniques hat use ML-based algorithms like CNN, artificial neura network, recurrent neural network, and deep learning are very few. But, these algorithms can be efficiently utilized as he steganalysis tools to analyze the hidden contents from he image [313]. They can capture the convoluted depen- dencies among the pixels to analyze and identify the pres- ence of secret bits. Therefore, these ML-based steganalysis echniques pose a big challenge to the heuristic or conven- ional ISTs. The conventional ISTs are well handcrafted, but they are heuristic. It means a certain level of exper- ise is required to understand where and how exactly to embed the secret bits. Conversely, with minimum effort, deep learning based CNNs can accomplish this. The only hing here is to design the structure of the encoder and de- coder. For example, the artificial neural network can effec- ively identify the optimal embedding location in an image. Therefore, in the near future, it is expected that the fight be- ween steganography and steganalysis is going to be the fight between the neural networks. Further, it is also ex- pected that the trained AI models can perform the embed- ding and extraction on their own when they are trained with supplying necessary features or inputs. Therefore, his can reduce human intervention for performing the em- bedding and extraction to a greater extent in the coming future. Therefore, ML-based techniques are promising to pay great attention to achieve a lot of wonders in the ste- ganalysis domain. Baluja [314] implemented LSBS and matrix coding- With the rapid growth in the field of artificial intelligence,
Table 14: ... continued 5 Steganography and steganalysis towards machine learning network can hide an image inside the OI to produce the CI, and the decoder network can produce the complete hid- den image from the CI. However, the robustness of the tech- nique has not been verified. In [315], a two-stage authen- tication based deep steganography and matrix encoding technique has been not directly embedd the distorti ing technig improve id entity au proposed. Here, the secret message is ed in the pixels of the OI. Therefore, on in the Cl is significantly less than other exist- ues. Primarily, the objective of this schema is to hentication. An invisible steganogra- phy technique associating generative adversarial network and mixed loss func ion has been suggested in [316]. Here, a secret image is embedded in the original color image of equal size. However, to obtain imperceptible CI, the se- cret bits are embedded only in the Y channel, which do not carry any color-related features of the image. A U-Net CNN structure-based IST with two different networks such as hidden and extraction networks has been proposed in [317]. Using the hidden network, the sender embeds the se- cret image in an OI, and the receiver extracts the SI from the CI to obtain the SI. The resulting CI does not exhibit any visual cues to the outsider. towards machine learning With the rapid growth in the field of artificial intelligence, many steganographers move towards ML-based steganog- raphy and steganalysis [312]. It is too early to speak abou whether a trained model can be really helpful in regards to hiding and recovering information from the digital objec accurately. However, recently some researchers proved hat it is possible for machines to hide the secret message in a digital image. Until now, the count of such techniques hat use ML-based algorithms like CNN, artificial neura network, recurrent neural network, and deep learning are very few. But, these algorithms can be efficiently utilized as he steganalysis tools to analyze the hidden contents from he image [313]. They can capture the convoluted depen- dencies among the pixels to analyze and identify the pres- ence of secret bits. Therefore, these ML-based steganalysis echniques pose a big challenge to the heuristic or conven- ional ISTs. The conventional ISTs are well handcrafted, but they are heuristic. It means a certain level of exper- ise is required to understand where and how exactly to embed the secret bits. Conversely, with minimum effort, deep learning based CNNs can accomplish this. The only hing here is to design the structure of the encoder and de- coder. For example, the artificial neural network can effec- ively identify the optimal embedding location in an image. Therefore, in the near future, it is expected that the fight be- ween steganography and steganalysis is going to be the fight between the neural networks. Further, it is also ex- pected that the trained AI models can perform the embed- ding and extraction on their own when they are trained with supplying necessary features or inputs. Therefore, his can reduce human intervention for performing the em- bedding and extraction to a greater extent in the coming future. Therefore, ML-based techniques are promising to pay great attention to achieve a lot of wonders in the ste- ganalysis domain. Baluja [314] implemented LSBS and matrix coding- With the rapid growth in the field of artificial intelligence,
Related topics:
Computer Science
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved