Scribd
Upload
39 views

FExtract Result

The document discusses feature extraction in steganalysis. It explains that the accuracy of feature-based steganalyzers depends on the quality of the constructed feature spaces or image models. There are three types of steganalysis attacks: targeted, blind, and universal. Feature extraction performance is affected by image contents like backgrounds and edges. Background images have lower error and faster extraction times than high-detail images containing many edges. Errors in extracted text can include multiple spaces, new lines, and random character insertions.

Uploaded by

Tanveer Akhtar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as RTF, PDF, TXT or read online on Scribd
39 views

FExtract Result

The document discusses feature extraction in steganalysis. It explains that the accuracy of feature-based steganalyzers depends on the quality of the constructed feature spaces or image models. There are three types of steganalysis attacks: targeted, blind, and universal. Feature extraction performance is affected by image contents like backgrounds and edges. Background images have lower error and faster extraction times than high-detail images containing many edges. Errors in extracted text can include multiple spaces, new lines, and random character insertions.

Uploaded by

Tanveer Akhtar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as RTF, PDF, TXT or read online on Scribd
You are on page 1/ 6

· Feature Extraction

The accuracy of feature-based steganalyzers is inevitably linked to the quality of

constructed feature spaces (image models). Understanding the principles for design of feature

spaces and their intrinsic properties is therefore of a great importance for steganalysts’ success.

The never-ending battle between steganography and steganalysis could be seen as a

competition for a better model – the user tries to embed undetectably within her model while

the attacker tries to identify those relationships among cover coefficients that have been

neglected by the user and incorporates them into a new model. Developments in

steganography and steganalysis are thus strongly interrelated and the modern feature space

design reflects the successes and failures of both.

Steganalysis is widely known to have one of three attack types. The targeted steganalysis

attack scenario is when the attacker attempts to detect a specific steganographic algorithm

employed between the two users. This is commonly known as targeted attack, as the attack is

targeted to the given stegosystem. A small number of features may be sufficient for a

satisfactory detection performance. Such features, however, would unlikely perform well on a

different steganographic scheme. The blind steganalysis attack scenario is when the attacker

goal is to design a feature set that would be capable of detecting a wide range of

steganographic algorithms. The universal steganalysis attack scenario is another term that often

appears in the literature. However, the universal steganalysis attack scenario goal is to construct

a detector that would be able to detect all possible stegosystems, including those previously

unseen.
· Mean-Squared Error (MSE)

This section will illustrates the effect of background and edges in images. As a concept the

NN detect the pixels variations as embedded hidden message. When detecting the hidden

messages near the edges, the edges may be miss analyzed as ASCII characters that leads to

errors. Figure 515 shows low mean squared error for background images of 0.33 at epoch 0,

which means that it needs lower time and process capability to reach the final results.

Figure 515: Low mean squared error for the high background images.

On the other hand, the high edges images shows high mean squared error of 0.57 at

epoch 3 in Figure 516. High edges images needs higher time and process capability to reach the

final results.
Figure 516: High mean squared error for the high edges images.

Moreover, edges images gives higher errors compared to background images as shown in

Figure 517 and Figure 518. Edges gives a peak of 17 while the background shows a peak of 13

instances.

Figure 517: background images histogram showing low errors.


Figure 518: edges images histogram showing high errors.

· Encoding Time

As mentioned in the previous section, the background images requires less time and

processing capability to reach its final results compared to the high details images (i.e. edges).

Figure 519 shows the comparison between the (a) background image extraction time of 3

seconds and (b) edges images extraction time of 7 seconds.

(a) background images


(b) edges images

Figure 519: extracting time comparision between background and edge images

· Errors Types

In order to find the error types, we compare the test text file before embedding as shown

in Figure 520 and after extracting as shown in Figure 521.

Figure 520: text text before embedding


Figure 521: text text after extration

As shown from the previous two figures, the error in test text can be classified in one of

three types. These errors are multiple spaces, new lines and insertion of random characters like

“ ï»? ”. The multiple spaces are due to the wide background around the embedded characters.

The new lines are due to randomization of embedding characters positions within the image.

The random characters insertion are due to miss edge estimation which is incepted as error

within the extracted text file. However, the proposed algorithm is immune to missing of

character error and the out of order error.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy