RESEARCH ON IMAGE DIGITAL WATERMARKING OPTIMIZATION ALGORITHM UNDER VIRTUAL REALITY TECHNOLOGY

. Aiming at the shortcomings of current algorithms due to the ﬁxed steps, which is easy to fall into local optimum, with robustness and poor trans-parency, and cannot be balanced against various common attacks, an optimization algorithm of digital image watermarking algorithm based on Drosophila was proposed. In the support of the virtual reality technology, the original color host image was transformed from the RGB space to YCrCb space, and the pixel block of the Y component was divided into a certain size; according to the principle of forming DC coeﬃcients in the DCT domain, the DC coeﬃcient of each block is calculated directly in the airspace, and the amount of modiﬁca-tion for each DC coeﬃcient is determined based on the watermark information and the quantization step size; according to the distribution characteristics of DC coeﬃcient, watermarks are embedded directly in the airspace; the type of digital watermarking and digital watermarking pretreatment methods were determined by using Drosophila optimization algorithm. At the same time, digital watermark embedding, extraction rules and initial steps were selected and identiﬁed. The Drosophila optimization algorithm with step size reduces the balance between global and local search ability, which makes up for the shortcomings of traditional algorithm. The experimental results showed that the proposed algorithm can eﬀectively balance the invisibility and robustness of the watermark, and can resist all kinds of common attacks, which with a better visual extraction eﬀect.

1. Introduction.Virtual reality technology is a computer simulation system that can create and experience the virtual world, it is the use of computer to generate a simulation environment and a multi-source information fusion, interactive threedimensional dynamic visual and physical behavior of the system simulation to let the users immersed in the environment.With the rapid development of computer technology and virtual reality technology, digital products have a large spread in the network, and bring great convenience to people's work, study and life.However, because of its easy access and copied makes the copyright protection becomes more and more difficult, resulting in piracy and infringement problems are becoming more and more serious.Under such circumstances, people urgently need a reliable 1428 YI ZHANG AND XIAO-LI MA technology to solve these problems.Digital watermarking technology has become the main choice and means for people to deal with such problems [2,9].
Digital watermarking technology makes use of the characteristics of human visual and auditory systems, adding certain meaningful information and watermarks into digital works such as images, texts, audio and video, that make it hard to distinguish the difference between the digital work after watermarking and the original digital work, and through special inspection method which can extract the information, in order to prove the copyright of the digital media.The watermark is closely combined and hidden in the original data, and become an inseparable part of it [4,10].Scientific research showed that 80% of the information that a person receives is from vision.Compared with text and audio information, the information content of digital image and video is larger and more intuitive, which plays a greater role in information dissemination, and also has greater commercial value in the market [16,18].In the past ten years, digital watermarking technology has made great progress, the traditional watermarking algorithms are basically for gray image, and color image watermarking algorithm has not been adequately studied, this is mainly because the gray image is easy to process, it only contains the luminance information, but not contains the color image information, in which the embedded watermark does not produce new color components.However, the color image is more common in real life and the dissemination of information on the Internet, which is more vulnerable to infringement, counterfeiting, abuse and with other bad behaviors.In recent years, the events such as "Southern China tiger incident", "CNN distorted reports of riots in Tibet", "Liu antelope", "Zhang Feige" are all related to color images.Therefore, whether the color image is used as the host image or digital watermark, color image watermarking technology attracted more and more attention and become one of the hot spots of image digital watermarking technology [5,15].
In order to solve the specific problem of Drosophila optimization algorithm, this paper proposes an image digital watermarking algorithm based on Drosophila optimization algorithm from two aspects: invisibility and transparency.The feasibility and objectivity of the proposed algorithm was proved by simulation experiments, which balanced the contradiction between robustness and invisibility of traditional algorithm effectively, and has strong anti attack ability against several common attacks.

2.
Research on image digital watermarking optimization algorithm under virtual reality technology.

2.1.
Blind watermarking algorithm for color images based on space and transform domain under virtual reality technology.Under the support of virtual reality technology, a blind watermarking algorithm based on the combining advantages of space and transform domain is proposed.First of all, the original color host image from the RGB color space is transformed to YCrCb color space, and the Y component was divided into the pixel blocks of 8 × 8; then, according to the DC coefficients forming principle in the DCT domain, the DC coefficient of each block in the airspace was calculated directly, and the amount of each modification the DC coefficients was determined according to the watermark information and the quantization step; finally, according to the distribution characteristics of the DC coefficient modified amount, watermarks are embedded directly in the airspace.The specific steps are described as follows: (1) DC coefficient of the image obtained in the airspace under the virtual reality technology Assumed that an image of the size of M × N in the virtual reality technology is expressed as f (x, y)x = 0, 1, 2, . . ., M − 1, y = 0, 1, 2, . . ., N − 1, and the DCT transformation formula is as follows: Where: In the same way, the expression of the DCT inverse of the image f (x, y) is as follows: According to the formula (1), it is known that, when u = 0 and v = 0, the image DC coefficient C(0, 0) in the DCT domain can be described as the following form: According to the above formula, we can see that the DC coefficient C(0, 0) of the image can be obtained through simple mathematical operations in the airspace without the need of complex DCT transformation, which can reduce the computation time of cosine or cosine.
(2) The feasibility of embedding watermark by using the obtained image DC coefficient Generally, the process of watermark embedding in DCT domain is actually to add watermark information to DCT transform coefficients, and then getting the watermark image through inverse DCT transform [11,14].In the following, it is feasible to embed the watermark in the DC coefficient from the viewpoint of energy conservation.
Assumed that a foreign watermark signal is expressed as E(i, j) and an arbitrary coefficient C(i, j) after DCT transformation, where i = 0, 1, 2, . . ., M − 1 and j = 0, 1, 2, . . ., N − 1 are changed to C(i, j) • , and the transformation formula is as follows: The modified image f (x, y) • can be calculated by inverse DCT transform.The calculation formula is as follows: Where, e(i, j) represents the signal that the image added to the (i, j) DCT coefficient of the (x, y) pixel block under the virtual reality technology.The energy of the added signal is calculated as: According to the above formula, we can see that in the DCT domain, the modification amount of any main coefficient is the same as that of the inverse transformation in the airspace, and embedding watermark is also no exception for the DC component of the image.
(3) The DC coefficient modifying in the image space under the virtual reality technology In the result of image DCT transformation, except for the DC coefficient, the rest are AC coefficients.Therefore, the DCT inversion described by formula (3) can be rewritten as follows.
Where, f (x, y) AC represents an AC component image that is reorganized by AC component data in virtual reality technology.
The main image block after the Harajuku hypothesis can be expressed as: Where, m and n are the coordinates of pixel points in each pixel block under virtual reality technology, and the original images are segmented into non overlapping blocks of size b × b.
The above formulas ( 10) and ( 11) can be further derived as follows: The above formula showed that for the host image f (x, y), embedding watermark in the DCT domain by DC coefficient can be directly implemented in the airspace, that is, adding ∆M/b to every pixel in the b × b block can embed the watermark in the airspace.Generally, a complete watermarking algorithm will include three processes, which are watermark generation, watermark embedding and watermark extraction [1,12].When generating watermark, the watermark algorithm in this chapter uses hash scrambling to improve its security and robustness [8,17].
In the stage of watermark embedding and extraction, the method of coefficient quantization is used to achieve the purpose of blind extraction, and the detailed steps of the algorithm are described as follows.
(1) Generation of image digital watermarking Assuming that the original watermark is shown in Figure 1 (a), the result of hash scrambling based on key K1 is shown in Figure 1 (b), which will further improve the robustness and security of image watermarking.
(2) Digital watermarking embedding in virtual reality technology (a) The conversion of the host image from the RGB space to the YCrCb color space; (b) The Y component in the YCrCb color space is obtained and divided into a non overlapping pixel block of 8 × 8. (c) According to the above formula (4), the DC coefficient C i,j (0, 0) in each sub block is calculated directly in the airspace.(d) The image DC coefficients were quantized according to the quantization tables QA(k) and QB(k), in which the quantization table is based on the quantization step of the key K2.The calculation formulas are as follows: Where, min(•) and max(•) respectively represent the minimum and maximum values of the image quantization watermarking coefficient, round(•) represents the integral function.
(e) According to formulas ( 16) and ( 17), the amount of modification M C i,j of the DC coefficient of the image is calculated.
(f) According to formula (17), and M C i,j /8 value was added on all pixels of the image, that is, a watermark bit was embed into the pixel block in the spatial domain.

(3) Image watermark extraction under virtual reality technology
The following steps are performed on the premise that the original host image or the original watermark image is not needed, and the following steps are performed to extract the watermark.(a) The watermark image I • is converted from RGB to YCrCb color space under the virtual reality technology.(b) The Y component of the YCrCb color space is obtained and divided into a non overlapping pixel block of 8 × 8. (c) The DC coefficient C i,j (0, 0) of the image is obtained directly by the above formula ( 4).(d) According to formula (18), the hash converse transformation is performed on the image w(i, j) • based on the key K2, and the final extraction of the watermark W • is obtained.
In the formula, mod( • ) represents the complementary function; ceil(x) represents the smallest integer that is not less than x.(e) Using key K1 to perform hash converse transformation on image w(i, j) • , and the final extraction of watermark W • was obtained.

Image digital watermarking optimization algorithm based on
Drosophila optimization algorithm.Drosophila optimization algorithm of digital image watermarking algorithm is an adaptive image digital watermarking algorithm, the algorithm determines the type of digital watermark used in the digital watermarking algorithm and the digital watermark preprocessing, and the digital watermark embedding rules and initial steps are extracted and determined [6,19].Drosophila optimization algorithm is a new method based on Drosophila foraging behavior to find global optimum.The olfactory organ of Drosophila can collect all kinds of smell in the air, fly near the food location, find the partners to gather positions by vision, and fly to this direction.The food searching is a process of repeated iterations.Drosophila need to repeat the location of food odors and peer positions, change their location repeatedly, and find the best food location in the global range [3,20].
According to the characteristics of the Drosophila searching for food, the following mathematical models are used to simulate it.
Step 1. the size of the Drosophila population is assumed to be Size, and the maximum number of searches for food is expressed as L max , and the expression of the group location for the initialization of the Drosophila flies is as follows: Step 2. Drosophila individuals through smell to find the food direction (X i , Y i ) and the distance R, the expression is as follows: Step 3. find the individual position of the Drosophila with the highest food taste concentration in the Drosophila population, that is to find the best individual, and the calculation formula is as follows: Step 4. iterative optimization, repeat the above steps 2∼5 to control the number of iterations of Drosophila melanogaster population by L max , the food searching range of each iteration determined by the Size, and determine if the concentration of flavor food taste is better than that of the previous iteration, if it is, go to step 4.
The optimization algorithm of Drosophila has the largest search step in the early stage of the iteration.At this time, the Drosophila individual has the largest foraging range and the global search ability is the strongest [7,13].With the increase of the number of iterations, the foraging range of the Drosophila gradually decreased, the global search ability gradually decreased, and the local search ability was gradually enhanced.
(1) Assuming that the size of Drosophila population is expressed as Size = 110, the largest number of search food (i.e. the maximum number of iterations of population) is expressed as L max = 5, the location is X i , Y i ∈ [0, 10] and initial step of Drosophila group is R 0 = 4.
(2) The food taste concentration of the Drosophila individual.The food taste concentration can be obtained by replacing the food taste concentration determination value S i into the taste concentration determination function f unction.The calculation formula is as follows: Where, P SN R i and N C i represent the peak signal to noise ratio and similarity of the carrier images embedded in the watermark.PSNR is based on the difference between the original vector image and the modified watermarked image and reflects the quality of the watermarked image under the virtual reality technology; the NC value is between the original watermark and the extracted watermark similarity metrics, if the NC value is more close to 1, they are more similar.The expressions of PSNR and NC are as follows: Where, P (i, j) represents the original image watermark, and P r (i, j) is used to extract the watermark.
The value of PSNR is in decibels (dB).In general, the larger the PSNR, the less the watermark embedding is caused by the quality of the carrier image, and the better the invisibility of the watermark image.Subjectively, the tolerable image PSNR value is on the top.According to the visual characteristics of human eyes, when P SN R ≥ 28dB, it considers that the invisibility of the watermarking algorithm is better, the more the NC value is closer to 1, the better the image watermark is obtained.When the degree of attack is certain, the higher the quality of the image watermark extraction, the stronger the robustness of the watermarking optimization algorithm is.
(3) Iterative optimization, repeat step 2 to 5, to control the number of Drosophila population iterations by L max , the optimal range of food taste during each iteration is determined by the Size, that determined the concentration of flavor food taste is better than that of the previous iteration, if it is, go to step 6.
The Drosophila optimization algorithm with decreasing step size achieves the balance between global and local search ability, and makes up for the deficiency of traditional Drosophila algorithm due to fixed step size, so as to ensure that Drosophila has larger probability to find the global optimal solution, and the algorithm does not fall into local optimum.To sum up, Drosophila optimization algorithm is used to adjust and optimize the image digital watermarking algorithm based on virtual reality technology, and set up initialization step R 0 = 4, which can realize image digital watermark optimization.
The integer DCT transform used in this paper is an 8 × 8 block integer DCT transform.The image is divided into b × b non overlapping sub blocks, each sub block is C (i,j) .According to the following formula, the watermark information W • is embedded into the DC coefficient of C (i,j) block.
Where, max C (i,j) and min C (i,j) denote the largest and minimum values of the 64 integer DCT coefficients of the image block, respectively.S represents the embedding strength of the watermark; when the watermark information W • = 1 is embedded, the constant parameter ω = 1; when the watermark information W • = 0 of the image is embedded, the constant parameter ω = 0.
The DCT transformation is carried out to carry the constipation image, and the integer DCT transformation is selected by the image subgraph block.The watermark information is extracted in the sub block C (i,j) according to the following formula (26).
In the addition rule of watermark embedding in formula (27), the watermark embedding strength S has a great influence on the invisibility and robustness of the watermark.If the greater the embedding strength S is, then the DC coefficients of the watermarking embedding position changes greater, the quality of the carrier image is reduced, and even makes the use of digital watermarking meaningless; if the embedding strength S is smaller, the DC coefficient changes smaller, and the watermarking information is not easy to detect, good invisibility, but if the secret image is attacked, the watermark is not easy to extract, that is, the robustness of  According to the above analysis, by using Drosophila optimization algorithm, the embedding strength S got adaptive adjustment and optimization through the P SN R i and N C i value, as far as possible to improve the image of the P SN R i and N C i values, ensure that the watermark has good invisibility, robustness and higher recognition rate, while ensuring the watermark embedding of the carrier image change little, that does not affect the use of carrier image.

Experimental results and analysis.
3.1.Experimental environment and parameter setting.In order to prove the validity and feasibility of the proposed algorithm, the virtual reality technology is used in the simulation experiment under the MATLAB software environment.Where, the maximum iteration number of Drosophila algorithm L max is set to 1000, the population scale Size is set to 20, the search step length R is set to 0.05, and the initial step length R 0 is set to 4. Simulation experiments using Baidu random search 256×256 color image as a carrier image digital watermark embedding, as shown in Figure 2. The watermark image is a meaningful watermark image with a "" word, as shown in Figure 3 (a).The number of watermark scrambling times is set to 16 by using virtual reality technology, and the watermark image after the scrambling is shown as Figure 3 (b).At the same time, the PSNR value and NC value of the above 2.3 sections are used to evaluate the invisibility of the image watermark and the robustness of the algorithm.
In the simulation experiment, the following five kinds of attacks are selected as the performance indexes to evaluate the effectiveness of the proposed algorithm: the number of attack experiments is 5, and the value of N C i comes from the following six kinds of attacks at once: 1 salt and pepper noise with intensity of 0.02; 2 5 × 5 template median filter; 3 counterclockwise rotation of 45 degrees; 4 first reduced to half of the original and then enlarged to the size of the original image; 5 JPEG compression with a quality factor of 90; 6 image cropping.After many simulation   experiments, the use of Drosophila optimization algorithm iterative optimization L max has not reached the maximum number of iterations before the iteration termination condition has been reached, that is, the exit flag exit flag equal to 0, to determine the current optimal embedded intensity value, the watermark image is obtained as shown in Figure 4.
From the visual observation, we can see that the two images in Figure 4 (a) and Figure 4 (b) are basically the same as the watermarked images, without causing the visual difference of the human eye and reaching the invisible watermark Claim.

Anti-attack experiments and subjective evaluation results analysis.
Six attacks were performed on the two watermarked color images of Figure 4, and the robustness of the algorithm was evaluated by simulation results.The attacked watermarked image and corresponding watermark extraction results are shown in Figure 5.    5. Observe the data in Table 2, the NC values are above 0.95, which showed that the proposed algorithm has good robustness to common attacks, especially the NC value under JPEG compression and cropping attack has a good performance.
To further illustrate the robustness of the proposed algorithm, the two watermarked images in Figure 4 continue to be subjected to higher saltiness noise and JPEG compression attacks.The experimental results are shown in Table 3.
As can be seen in Table 3, for the decrease quality of JPEG compression, the more aggressive attack, the smaller the NC value, and the weaker the visual effect of extracting the watermark.A JPEG compression NC with a quality factor of 30 can be maintained at above 0.93, which can extract watermarks with good visual effects.For salt and pepper noise intensity increases, the more aggressive attack, the smaller the NC value, the weaker visual effect of watermarking will be.When the salt and pepper noise attack intensity of 0.20, NC value remained above 0.95, the extracted watermark will be recognized by the human eye, which does not affect the meaning of its express.In summary, this algorithm has good anti-attack performance for different levels of salt and pepper noise and JPEG compression.

4.
Conclusions.In this paper, the Drosophila optimization algorithm is used to adaptively determine the embedding strength of watermarking, which makes the watermark with a good concealment and improves the robustness of the algorithm.It avoids multiple comparison of experimental results in a simulation experiment to manually set the appropriate embedment intensity factor.For different vector images, the embedding intensity factor needs to be adjusted more, which also avoids the cumbersome adjustment of the embedding intensity factor due to the replacement of the carrier image.In summary, this algorithm effectively balances the invisibility and robustness of watermarking, which has some practical significance.
The experimental results showed that the embedded watermark can be extracted with higher accuracy, the watermark similarity detection is above 95%, which has strong robustness, especially for geometric attacks such as rotation and cropping.At the same time, Watermarked images are difficult to observe by human vision with good transparency.In addition, the watermark is pre-encrypted before embedding, which effectively ensures the security of the watermark.

Figure 1 .
Figure 1.Image digital watermarking preprocessing under virtual reality technology (a) Color images of flowers (b) Landscape color image

Figure 2 .
Figure 2. Color images of the original carrier

Figure 3 .
Figure 3. Original watermark image and scrambled watermark image

Figure 5 .
Figure 5. Watermarking effects and watermarking extraction effects after different attacks

Table 1 .
PSNR values of the original and watermarked images for each attack Objective evaluation results.The PSNR and NC values under six kinds of attacks will be calculated to objectively analyze the attack results.PSNR values have varying degrees of change, indicating watermarked images have different degrees of distortion, which is as the same as shown in Figure

Table 2 .
NC value of the original watermark and extracted watermark image under various attacks

Table 3 .
Experimental results of two watermarked images under varying degrees of attack