LSB-based pre-embedding video steganography with rotating & shifting poly-pattern block matrix

Fixed block structure

In algorithms using the fixed block structure, the same KBM is used for each 64 × 64 pixel block area of the video frame. The KBM includes pixel areas that are suitable for the data embedding as mentioned before.

After the completion of the data embedding process by using a KBM in the first 64 × 64 pixel block area of the video frame, the same KBM is used in the next 64 × 64 pixel block areas of the video frame. The processes continue until the whole data is embedded within the video frames.

Variable block structure

In algorithms using a variable block structure, the first placement of the KBM into the video frame is the same as in the fixed block structure. After the completion of the data embedding process performed by using a KBM in the first 64 × 64 pixel block area of the video frame, the KBM will be changed for the next 64 × 64 pixel block area of the video frame. This change is implemented according to the colour values of the most recent data-hidden pixel-which is the last appropriate pixel for the data embedding within the 64 × 64 pixel block area of the current video frame. Shifting, rotating, or shifting-rotating operations in the KBM depending on the most recent data-hidden pixel are carried out as follows:

• If all the RGB colour channels of the most recent secret data-hidden pixel are used to embed the data then B and G colour channel values of the related pixel and

• If only one channel of RGB colour channels of the most recent data-hidden pixel is used to embed the data then the values of colour channels in which no secret data is embedded will be used for changing the structure of the KBM.

Let us make clear the variable block structure by giving an example. The process steps for the determination of the next KBM structure (movement processes) are as follows for sample values:

Step 1. The colour values of the last and most recent secret data-hidden pixel within the current block are read. Assume that these values are B = 178, G = 205 and R = 114

Step 2. The first LSB is the bit where the data is embedded. The second LSB represents the control bit and the third, fourth, fifth, and sixth LSB values represent the number of steps for related operation. Assume that each RGB colour channel is used for the data embedding. The colour channels B (ch1) and G (ch2) are used as control and step indicators for left–right (B) and up–down (G) block movements, respectively.

 

B = ch1 = 10110010   G = ch2 = 11001101

 

For ch1 of pixel value,

 

178 = 10 1100 1 0

 

Embedded secret data (LSB) = 0

 

For control bit (ch1_control) = 1

 

For step number (ch1_step) = (1100)₂ = 12-pixels is the calculated value.

 

For ch2 of pixel value,

 

205 = 11 0011 0 1

 

Embedded secret data (LSB) = 1

 

For control bit (ch2_control) = 0

 

For step number (ch2_step) = (0011)₂ = 3-pixels is the calculated value.

Step 3. In the directions determined by the control parameter, the 16 sub-pattern blocks that make up the KBM are moved inside KBM as much as the calculated number of pixels. The reader is kindly requested to accept the KBM movement expression as the movement of the 16 sub-pattern blocks in the KBM from now on. As a result of these movements, the structure of the next KBM is determined.

For the shifting process:

 

For ch1_control = 0 or 1,

 

0 means shifting right process,

 

   1 means shifting left process.

 

ch1_step = number of steps in pixel regarding to shift right or left

 

For ch2_control = 0 or 1,

 

   0 means shifting up process,

 

   1 means shifting down process.

 

ch2_step = number of steps in pixel regarding to shifting up or down

For the rotating process:

 

For (ch1_control AND ch2_control) == 0 or 1 value,

 

   0 means rotating right process (90^∘)

 

   1 means rotating left process (90^∘)

 

ch1_step = number of steps in pixel regarding to rotating right

 

ch2_step = number of steps in pixel regarding to rotating left

 

Since the KBM will be equal to itself after every 4-steps rotation, the shifting process is implemented with modulo 4 operation.

 

   ch1_step = ch1_step % 4 or

 

   ch_2step = ch2_step % 4

For the shifting-rotating process:

 

Assuming that the movement selection for the determination of the KBM is ”shifting-rotating”, according to the sample values.

 

For the values of ch1_control = 1 and ch1_step = 12, the KBM will be shifted 12 steps in the pixel in the left direction.

 

For the values of ch2_control = 0 and ch2_step = 3, the KBM will be shifted 3 steps in the pixel in the up direction.

 

For (ch1_control and ch2_control) == 0

 

  ch1_step = 12

 

  ch1_step = ch1_step % 4 values, the KBM will be rotating 0 step in the right direction.

The process to be followed where only one channel of RGB is used for the data embedding includes similar steps for changing the KBM. In this case, the colour channels in which no secret data is embedded will be used as control and step indicators for left–right and up–down block movements.

Creation of the stego object – encoding

Figure 6 shows the flowchart displaying the creation of the stego object based on the proposed method. First, the user gives a name to the .txt file including the secret message to be hidden. Video recording is made for the cover object to be used in the creation of the stego object. The dynamically produced cover object is created through the user recording a momentary image with a camera. The video frames and audio files are recorded separately to be combined afterwards. Each video frame is given a number based on the order of recording in the folder and saved as .png while the audio file is saved as .wav. The user chooses the methods used to generate the stego object. First, the user selects the algorithm to be used in the data embedding process among algorithms namely the sequential data embedding, the sequential data embedding in all frames, and the non-sequential data embedding in all frames. Then, 16 sub-pattern blocks and type of KBM (fixed, shifting rotating, or shifting-rotating) are selected. Afterwards, the colour channel to embed the data is selected among RGB, R, G, or B. The ASCII data input by the user in the .txt (text) format is converted into the binary system. Then, the embedding process is implemented according to those selected parameters.

The steps of the process of embedding sample data within a dynamically recorded video are as follows:

Step 1. The secret message is entered.

Let’s assume that the message has a size of 42.8 kb.

Step 2. Dynamic video recording is performed as the cover object.

The dynamically recorded video has a pixel size of 640x480 and consists of 330 video frames.

Step 3. The selection of the control values is performed.

Assume that the selections are as follows:

 

The data embedding algorithm selection: “The Sequential Data Embedding” algorithm,

 

Block structure selection: The shifting-rotating block structure,

 

colour channel selection: The “RGB” colour channel.

Step 4. After the completion of the data input processes, the input message is converted into a binary code and the message length is calculated. The message, that is recorded as ASCII, is kept in a .txt file by being converted into a binary code. The sample message length is calculated as 376193 bits. This value is included in the control values.

Step 5. A KBM is created depending on the ordering of selected 16 sub-pattern blocks. This KBM is included in the control values.

Step 6. To obtain the secret message at the receiver side, the information regarding the KBM, colour channel, message length, and the data embedding algorithm are embedded within the video frames as control values.

Step 7. Based on these selections, the data embedding process is performed.

Figure 7 shows the 11th (left) and 12th (right) KBMs produced depending on the selection of KBMs structure and type (shifting-rotating block) for the 1st video frame. As can be seen from Fig. 7, 12th KBM is shifted and rotated version of 11th KBM. Figure 8 shows the areas where data was hidden in the 2nd video frame. The number of frames that constitute the video and their resolution values are used to calculate the maximum capacity of data to be hidden in the video.

The data to be hidden are placed in the fields obtained by dividing the pixels of the video frame into 64 × 64 bit blocks. The last block fields parsed for each frame are used to hide control data. The number of blocks in which data can be hidden within the video frame is determined as follows:

For a (w) x (h) video:

where ’w’ and ’h’ represents width and height of the video frame in pixel, respectively. The last 64 × 64 bit block areas of the frames where data can be hidden are used as areas reserved for hiding the control values. In this case, the total number of blocks to hide data for each video frame is: (4)${\displaystyle tnb\text{\_}final=tnb-1}$

The number of blocks in which data can be hidden for each frame of a sample 640 × 480 p video is calculated as follows:

 

nhb = int(640/64) = > nhb = 10

 

  nvb = int(480/64) = > nvb = 7

 

  tnb = 10x7 = 70blocks.

 

  tnb_final = 70 −  − 1blockperframe.

Maximum number of bytes that can be embedded into the cover object regarding to the usage of last bit of each channel of an RGB pixel; (5)${\displaystyle Capacity\left(RGB\right)=\left(numberofframesxtnbx64x64x3\right)/8\left(bytes\right)}$

Maximum number of bytes that can be embedded into the cover object regarding to the usage of only one colour channel of the pixel of an RGB image; (6)${\displaystyle Capacity\left(R,GorB\right)=\left(numberofframesxtnbx64x64\right)/8\left(bytes\right)}$

The capacity of the video file must be at least twice that of the secret message since KBM has black regions that correspond to pixel in which no data can be embedded. The proposed approach aims to embed secret text messages into the video. When the user would like to embed video or audio messages by the proposed approach, the related file format must be converted into binary format. Later on, depending on the length of the converted file (bytes), the cover video must be shot. The length of the cover video t in time (second) can be calculated for a (wxh) video by using Eqs. (6) and (7). (7)${\displaystyle t=convertedfilelength\left(bytes\right)/Capacity\left(R,GorB\right)}$

The sizes of encoded and decoded secret messages are presented in Table 2. This table shows that the proposed method is a lossless video steganography method. Table 3 includes the times it takes for encoding and decoding each of secret message with respect to 3 different algorithms. Considering the above equations, it should be noted that as the size of the secret data presented in Table 2 grows, the encoding process that is calculated by Eq. (7) takes longer as can be seen in Table 3. Thus, depending on the time-cost efficiency, power of the computer and the requirements, the user is free to choose the secret message type (video, music or text) and size.

Obtaining the secret message – decoding

For the decoding operation, the user is initially asked to enter the name of the stego object in the .avi file format as the input parameter. The input stego object is divided into frames. As the next step, the control values hidden within the last 64 × 64 pixel block areas of the video frames are obtained. The control values are read according to the order of embedding in the video frames. The information regarding the selected colour channel, the selected message’s length, the selected structure and type of KBM, and the selected algorithm is extracted. Finally, the secret data is extracted sequentially from the first video frame using control values and KBM. The LSBs of the related RGB colour channels of each pixel in which the data is embedded are read. All the data is saved in a ”.txt” file in binary format that will be converted into ASCII characters in the same file.

MSE

MSE is the mean of the squared difference between the corresponding pixel values of two images. It is an image quality evaluation based on the error sensitivity between the corresponding pixels. It gets an error value between 0 and 65025 for the 8-bit colour value (Shahid et al., 2015). This value is expected to be as small as possible for a high-quality video processing result. A value of 0 states that there is no difference between the two videos while 65025 represents the maximum difference for the 8-bit colour value. The closer the result is to 0, the more successful the video processing result is considered to be. The MSE value is calculated based on the following equation Eq. (8). (8)${\displaystyle MSE=\frac{1}{n.m}\sum _{x=1}^n\sum _{y=1}^m{\left(|f_{\left(x,y\right)}-f_{\left(x,y\right)}^{{}^{\prime }}|\right)}^2}$

where, m and n are the rows and columns of the cover image respectively, and f_(x,y) and $f_{\left(x,y\right)}^{{}^{\prime }}$ means the pixel value at position (x,y) in the cover-image and the corresponding stego-image, respectively.

PSNR

PSNR is defined as the ratio between the maximum possible power of a signal and the noticeable power of the potential distortion (noise level in the video) in the processing and transmission of video. The logarithmic decibel scale (dB) is generally used for PSNR due to the wide dynamic range of signals. If the calculated PSNR value is close to 1, it indicates that the quality of similarity between two images is high. The PSNR value is calculated based on the following equation Eq. (9) (Haas et al., 2019). (9)${\displaystyle PSNR=10.{log}_{10}\left(\frac{{\left(f_x\right)}^2}{MSE}\right)}$

where f_x represents the value of maximum pixel or maximum signal in the image. It is calculated as f_x = 2ⁿ − 1 for the maximum signal value with n representing the pixel bit value. For example, pixel values are calculated as f_x = 255 for 8-bit colour values and as f_x = 1023 for 10-bit colour values.

SSIM

SSIM is a metric that displays the structural similarity between two images. In calculation, it uses structural distortion, perceptual structure and differences in luminance values. SSIM performs a comparison of the structural similarity between the original image and the processed image. The result of the comparison gives a value between 0 and 1. The quality and the similarity between the images increase as the result gets closer to 1. In the analysis process, SSIM takes local luminance and contrast into consideration. SSIM measures contrast, luminance and structural comparison using mean and variance values (Vranješ, Rimac-Drlje & Grgić, 2013).

For the SSIM value;

Luminance, contrast, and structural comparison functions are calculated based on the following equations, respectively. (10)${\displaystyle I\left(x,y\right)=\frac{2{\mu }_x{\mu }_y+C_1}{{\mu }_x^2+{\mu }_y^2+C_1^{{}^{\prime }}}}$

The luminance comparison function l(x, y) is a function of μ_x and μ_y.μ_x and μ_y represents luminance of x and y signals and are estimated as the mean intensities. Constant C₁ is included to avoid instability when ${\mu }_x^2+{\mu }_y^2$ is very close to zero. (11)${\displaystyle c\left(x,y\right)=\frac{2{\sigma }_x{\sigma }_y+C_2}{{\sigma }_x^2+{\sigma }_y^2+C_2^{{}^{\prime }}}}$

The contrast comparison c(x, y) is the comparison of σ_x and σ_y. The standard deviation as an estimate of the signal contrast is used. But, in contrast comparison an unbiased estimate in discrete form (σ_x, σ_y) is used. C₂ is a non-negative constant. (12)${\displaystyle s\left(x,y\right)=\frac{{\sigma }_{xy}+C_3}{{\sigma }_x{\sigma }_y+C_3^{{}^{\prime }}}}$

Structure comparison is conducted after luminance subtraction and contrast normalization. As in the luminance and contrast measures, a small constant C₃ in both denominator and numerator is introduced.

When the three comparisons represented by Eqs. (10), (11) and (12) are combined to produce the similarity metric (SSIM) obtained from the signals x and y, the following Eq. (13) is obtained. (13)${\displaystyle SSIM\left(x,y\right)={\left[I\left(x,y\right)\right]}^{\alpha }{\left[c\left(x,y\right)\right]}^{\beta }{\left[s\left(x,y\right)\right]}^{\gamma }}$

where α > 0, β > 0 and γ > 0 are parameters used to adjust the relative importance of the three components.