Simultaneous optimization of the resistance and transparency of image encryption based on psycho-visual techniques

Dissertation for M.Sc degree

Trend: Software Engineering

Abstract

The process of inserting information into a multimedia signal so that the effect of insertion is not obvious and the information can be extracted when needed is called encryption. To implement a cryptography idea, one should focus on the features of power[1], reliability and reliability[2] and invisibility[3]. Most of the existing methods of hiding color images are designed in such a way that inserting the hiding image is done only in the brightness component of the image. The most important disadvantages of these methods are: 1) These methods are sensitive to color attacks because they ignore the communication between different color channels. 2) Due to the lack of detection of inconsistency in the image, these methods are not resistant to geometric attacks. Therefore, the main challenge is to design a color image encryption idea that is resistant to geometric attacks. For this reason, by using quadruple Fourier transform, singular value matrices [4] and least squares support vector machine [5], we have proposed a blind and robust color image hiding idea in the field of quadruple Fourier transform, which has good visual quality. In this idea, first the original color image is divided into sub-blocks and then, on each of the sub-blocks, the fast fourier transform is performed. Finally, the digital image is inserted by adaptive modulation in the individual values ??of the real coefficients of the quadruple Fourier transform of the image blocks. In order to extract the latent image, according to the values ??of the pseudo-instantaneous moments (primary rows of the matrix), using the least squares support vector machine algorithm, correction is done on the latent image. The results of the simulations show that the proposed idea of ??hiding in color images, in addition to resisting common image processing operations such as adding noise, filtering and JPEG compression, is also resistant to geometric distortion.

Key words: image hiding, quad Fourier transform, singular value matrix, LS-SVM, geometric attacks

Chapter 1: general research

In this chapter, the concept of hiding first Digital images and their importance are discussed and then the objectives of the plan are explained. In the following, the research questions and hypotheses and innovations of the proposed algorithm are stated and explanations are provided about the keywords of the research. At the end, the structure of the plan is mentioned.

1-1- Introduction

Today, with the expansion of the Internet and computer networks, the issue of unauthorized copying of digital multimedia contents (such as image, video, audio, etc.) has become a more serious challenge. In this regard, digital encryption has been introduced as the best solution to detect the embedding, reproduction and unauthorized distribution of these multimedia contents (Cox and Miller[6], 2001). Encryption makes it impossible to access the desired product without knowing the correct encryption key, but after decoding, it is easy to make unauthorized changes to it or to reproduce and distribute it illegally. To enforce copyright laws and prevent unauthorized reproduction, interpolation and distribution of multimedia, the use of encryption is suggested as the best solution (Patra et al.[7], 2010).

Digital encryption is a method of including copyright information or other information in the original multimedia data. In other words, the insertion of information in a multimedia signal in such a way that the effect of the insertion is not evident and that information can be retrieved when necessary is called encryption (Ebrahimi Moghadam and Nemati [8], 2013). In encryption, the signal quality of the host should not drop more than a certain amount. In a cryptography system with copyright law enforcement applications, the following requirements must be met (Kogianos et al. [9], 2009):

1- Transparency[10]: The inclusion of the cryptography should not significantly affect the quality of the original signal. In fact, the latent image should be unrecognizable (invisible and inaudible) from the point of view of the human perceptual system.

2- Resistance[11] against attacks: intruders should not be able to delete or change the latent image, therefore, the latent image should be resistant to normal signal processing changes such as filtering, compression, etc.

3- The size of the thumbnail[12]: the number of bits that can be included in a specific number of host signal samples.

4- Security[13]: This means that the detection of the latent image is possible only through the authorized user.

5- The latent image can be extracted without the need for the original signal.

6- The latent image cannot be recovered without the knowledge of how to insert it.

7- It should be possible to insert the latent image directly into the signal, not the latent image only at the beginning of the signal.

Usually, most of these requirements are in conflict with each other and a compromise should be made according to the application, for example, increasing the encryption rate (the size of the image) causes a decrease in the quality of the host signal, so it is possible to increase the encryption rate to a certain extent so that the quality of the host signal is acceptable (Lin et al. [14], 2011). In most cryptographic applications, it is very important to provide the characteristics of transparency and resistance, and the conflict between these characteristics causes a fundamental challenge in the design of robust cryptography systems. Of course, most researches and applications are focused on resistant cryptography. Generally, due to resistance to most signal processing operations, resistant cryptography systems are used for copyright protection and ownership verification applications (Rahman et al. [17], 2011). 1-1-1- General framework of cryptography systems. Data (cryptography) in an image, audio, text or video signal is called cryptography. The general framework and procedure of a cryptography system is shown in Figure (1-2). In any cryptographic system, when necessary, it is possible to extract the cryptographic image from the encrypted signal. The thumbnail may contain data such as copyright information, license number, patent number, etc. be Signing paintings is a simple example of cryptography to prove ownership. Each encryption algorithm consists of three parts (Wu and Sun[18], 2013):

A- The cipher which contains unique information of the owner of the signal.

B-Encoder- to insert the cipher into the host signal.

1-1-2- types of cryptography

     According to the type of signal, human understanding, field of work and extraction procedure, cryptographic systems are divided into different categories, which are listed below for each of these divisions. According to the type of host signal, encryption systems are divided into the following four categories (Kogianos et al., 2009):

Image encryption

Audio encryption

Video encryption

Text encryption

From the point of view of human perception, encryption systems are divided into three categories:

Overt concealment

Covert concealment

Dual concealment

are divided (Lin et al., 2011). In open encryption, encryption is seen as opaque or even transparent in the host signal, open encryption is used in copyright protection and ownership proof applications (Khodai and Faez [19], 2010). While in hidden encryption, in the insertion procedure, the values ??of the host signal samples are changed in such a way that the effect cannot be understood by humans. Hidden cryptography is used as a witness and evidence for the ownership of the signal and also to detect unauthorized change in it (Criver et al. [20], 1998). Double cryptography is also a combination of overt and covert cryptography, where the hidden cryptography is used as the supporting information of the overt cryptography. Encrypting systems are divided into two categories based on the working domain:

space (time) domain encoding methods

frequency domain encoding methods

are divided (Patra et al., 2010). In the location domain stealth methods, the stealth information is inserted directly into the host signal samples, but in the frequency domain methods, after the signal is converted to the desired frequency domain, the stealth information is inserted into the frequency coefficients. In general, the frequency domain cryptography methods have much more resistance than the methods based on the time domain.

According to how the cryptographic extraction procedure works, these systems can be divided into three categories:

Seeing[21]

Pseudo[22]

Blind[23]

(Lin et al., 2011). In Bina systems, the main host signal and cryptographic key are needed to extract the ciphertext