A New Video Steganography Scheme Based on Shi-Tomasi Corner Detector

Recent developments in the speed of the Internet and information technology have made the rapid exchange of multimedia information possible. However, these developments in technology lead to violations of information security and private information. Digital steganography provides the ability to protect private information that has become essential in the current Internet age. Among all digital media, digital video has become of interest to many researchers due to its high capacity for hiding sensitive data. Numerous video steganography methods have recently been proposed to prevent secret data from being stolen. Nevertheless, these methods have multiple issues related to visual imperceptibly, robustness, and embedding capacity. To tackle these issues, this paper proposes a new approach to video steganography based on the corner point principle and LSBs algorithm. The proposed method first uses Shi-Tomasi algorithm to detect regions of corner points within the cover video frames. Then, it uses 4-LSBs algorithm to hide confidential data inside the identified corner points. Besides, before the embedding process, the proposed method encrypts confidential data using Arnold’s cat map method to boost the security level. Experimental results revealed that the proposed method is highly secure and highly invisible, in addition to its satisfactory robustness against Salt & Pepper noise, Speckle noise, and Gaussian noise attacks, which has an average Structural Similarity Index (SSIM) of more than 0.81. Moreover, the results showed that the proposed method outperforms state-of-the-art methods in terms of visual imperceptibility, which offers excellent peak signal-to-noise ratio (PSNR) of average 60.7 dB, maintaining excellent embedding capacity.


I. INTRODUCTION
Information security issues have increased dramatically in the past few years, as people began to worry about their data from being cracked over the Internet-for instance, piracy tracking, copyright protection, authenticity identification of digital works, and identity authentication. To address these issues, the science of steganography and cryptography has emerged [1]- [3]. Steganography is a science that takes video, image, audio, or other digital media as a medium and then conceals secret data into the medium through a particular algorithm. Whereas, cryptography is a science that converts a secret message into a meaningless form so that eavesdroppers cannot interpret it [4]- [7]. Although both steganography and cryptography attempt to protect data, the use of either one alone is not an ideal solution. Thus, sometimes, it is The associate editor coordinating the review of this manuscript and approving it for publication was Md. Asikuzzaman .  [12].
recommended that both approaches be integrated. In such a case, even if the attacker had doubts about the existence of the communication and managed to defeat the steganography technique, the attacker would still need to break the encrypted message to obtain the secret message [8]- [11]. Fig. 1 illustrates, in general, the steps involved in the embedding and extraction process of any steganography algorithm. The efficiency of any successful steganography VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ FIGURE 2. Three main categories of video steganography methods [24].
algorithms depends on four main factors: robustness, embedding capacity, security, and imperceptibility. Therefore, these factors should be taken into consideration when designing a new steganography algorithm as well as when improving the existing algorithms. Robustness refers to the resistance degree of the steganography algorithm against attacks and signal processing. Embedding capacity refers to the amount of data that can be embedded within the cover medium. Security refers to the inability of the attacker to extract the embedded data. Imperceptibility refers to the distortion degree in the original cover carrier due to the hiding process [13]- [15]. Compared to other digital media, digital video has more redundancy, providing a large capacity for hiding data. Besides, with the advent of the era of big data, a large amount of HD digital videos is transmitted over the Internet. Therefore, video steganography has attracted the attention of many researchers and has become a popular choice [12], [16]. Video steganography is the process of embedding a confidential message into a cover video. Where, it is used in many fields such as copyright protection, access control, medical systems, law enforcement [17]- [19].
In general, there are three main categories of video steganography methods, namely format-based methods, video codec-based methods, and still image-based methods [8], [20], [21]. Fig. 2 demonstrates these three different categories in a tree diagram. Still image-based methods transform video medium into frames and then apply methods of image steganography to the selected frames for data hiding purpose. These methods can be further categorized into two subcategories: transform domain methods and spatial domain methods. In the transform domain methods, the cover carrier is initially transformed into the frequency domain. Then, some coefficients of the frequency domain are selected to be replaced by the confidential message. Finally, the domain, with the altered coefficients, is converted back into the spatial domain. Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT), and Discrete Fourier Transform (DFT) are some examples of the transform domain methods. On the other hand, spatial domain methods directly hide the confidential message into the cover carrier. Least Significant Bit (LSB) techniques are the most popular techniques of spatial domain methods due to their low computational complexity and simplicity [8], [22]- [24]. LSB techniques operate by substituting the confidential message bits with some LSBs of pixels from the video carrier frames. The second category of video steganography is format-based techniques. These techniques are designed for a particular video format by taking advantage of the compression strategy and structure of the format. H.264/AVC is an example of format-based techniques [25]. The third category of video steganography is video codec-based techniques. These techniques attempt to take advantage of the 3D nature of videos and exploit the third dimension, which is the time dimension t in embedding. This extra dimension provides some additional features, such as motion components and motion vectors [24], [26].
This paper proposes a new approach for hiding information within digital videos, which is based mainly on two wellknown algorithms, namely Shi-Tomasi corner detector algorithm and Four Least Significant Bits (4-LSBs) algorithm. The proposed method first encrypts the secret information using Arnold's cat map algorithm. After that, it uses Shi-Tomasi algorithm to detect corner points regions in the Y (luminance) channel of each frame within the cover video. Finally, it embeds the secret information into each pixel of the detected regions using 4-LSBs algorithm.
The rest of the paper is structured as follows. Section II outlines some state-of-the-art methods related to video steganography. Section III explains Arnold's cat map algorithm and Shi-Tomasi corner detector as preliminaries. Section IV describes the proposed method in detail. Section V presents the experimental results with the discussion. Finally, section VI concludes the paper and suggests some future works.

II. RELATED WORKS
Due to the large capacity of digital videos for hiding sensitive data, video steganography has gained the attention of many researchers in the literature. This section reviews some recent video steganography methods that are closely linked to our proposed method.
M. Sadek et al. [27] presented a blind and robust approach to video steganography based on human skin areas of video frames. Their presented algorithm first detects the skin areas within each frame in the cover video to generate a skinmap. Then, it converts the skin map to a skin-block-map where skin pixels prone to error are ignored from the hiding process. Lastly, it applies a three-level of DWT on the blue and red color channels of each frame to embed the secret data within the coefficients of the identified skin pixels in the skin-block-map through quantization. Experimental results demonstrated that their proposed method achieves a high degree of imperceptibility.
K. Niu et al. [28] proposed a new reversible technique for video steganography using cover videos with H.264/AVC extension. They used Histogram Shifting (HS) of motion vector values to conceal the secret data within the identified reference frames of the cover video. In their proposed technique, the hidden information can be restored from the compressed cover video in a lossless format. The results revealed that their proposed algorithm achieves higher invisibility and capacity than other existing techniques in the literature.
K. Rajalakshmi and K. Mahesh [29] introduced a novel approach for video steganography, called Zero Level Binary Mapping (ZLBM). Their proposed method first converts the cover video into frames. After that, it utilizes Fuzzy Adaptive Median Filtering (FAMF) technique to exclude the impulse noise from the frames. Then, it employs the block-wise pixel grouping method to group the pixels within the refined frames. In the end, it embeds the secret data using ZLBM method and encodes it using patch wise code formation method. Experimental results demonstrated that their proposed method performs better than other related approaches in terms of PSNR rate.
Y. Liu et al. [30] suggested a new and robust technique for video steganography based on the H.265/High-efficiency video coding pattern. Before the embedding process, their presented method encrypts the confidential data using BCH code approach. To restrict the intra-frame deformation drift, they used three groups of the prediction directions. After that, they embedded the encrypted secret data into the multicoefficients of the selected four by four luminance discrete sine transform blocks, which meet the groups. Experimental results revealed that their proposed method obtains better visual quality than the methods studied earlier.
M. Hashem Zadeh [31] recommended an efficient approach for video steganography based on the salient and dynamic areas of a cover video. His algorithm identifies the dynamic regions based on motion clues of the feature points where the areas of interest are determined consequently. To conceal the secret data within the designated areas, he used the least-significant-bit substitution technique. Experimental results showed that his proposed method attains a higher hiding capacity when compared to the state-of-the-art approaches in the literature.
The major limitation of the existing video steganography methods in protecting sensitive data is the difficulty to guarantee a good trade-off between imperceptibility and robustness. Thus, this paper proposes a video steganography method that operates in the special domain using Shi-Tomasi corner detector algorithm and 4-LSBs algorithm to overcome the limitations of the pervious works. The reason for using corner detection points rather than other types of detection points is that its high robustness against different types of attacks and its ability to ensure high imperceptibility. Besides, it gives an acceptable embedding capacity. Furthermore, to increase the security level of the proposed method, Arnold's cat map algorithm is used before the embedding process.

III. PRELIMINARIES
This section explains Arnold's cat map algorithm and Shi-Tomasi corner detector algorithm in detail. This is due to the use of these two algorithms in the proposed method.

A. ARNOLD'S CAT MAP
Arnold's cat map, also referred to as Arnold's transformation, is one of the chaotic maps that operates in the discrete-time domain, which can be used only with square images. It was discovered by Vladimir Arnold in the 1960s using an image of a cat; hence the name came from. This method applies a transformation to an image so that it rearranges its pixels randomly. However, if iterated sufficient times, ultimately, the original image reappears. The number of considered iterations is known as Arnold's period. The period depends on the size of the image; for instance, images with different sizes may have different Arnold's period [32]- [34]. Arnold's transformation can be expressed mathematically as in Eq. (1).
where (x , y ) is the new pixel location, (x, y) is the original pixel location, and N is the size of the square image. Cat map has two exemplary features that bring chaotic movement, namely tension and fold. Tension refers to the process of multiplying a matrix by x, y to enlarge x, y. Whereas, fold refers to the process of bringing back x, y within the unit matrix by using ''mod N '' [34]- [37]. Eq. (1) is employed to apply Arnold's transformation to every pixel in the image. When all the pixels, in the image, are transformed, the resulting image will appear in a meaningless form. At certain steps of iterations, if the resulting image reaches the expected target (for instance, up to the secret key), the requested scrambled image will be obtained. The decryption of image depends on the transformation periods (for instance, the number of iterations to be followed = Arnold's period − secret key). Fig. 3 shows the results of a cat image whose size is 150 × 150 after being transformed by Arnold's transformation using different iteration steps.

B. SHI-TOMASI CORNER DETECTOR
Shi-Tomasi algorithm, also referred to as Good Features to Track, is one of the corner detection approaches that is widely used in the field of computer vision to select certain types of features from an image. It is an improved version of Harris corner detector [39], [40]. Therefore, we start by illustrating Harris corner detector, and then we highlight the improvement aspect of this algorithm.
Harris corner detector is an eigenvalue-based feature point detector, which is the most widespread corner detector due to its strong invariance to image noise and rotation. It is based on the local auto-correlation function of a signal which measures the local changes of the signal with patches shifted by a small amount in different directions [41]- [43]. Given a shift ( x, y) and a point (x, y), the auto-correlation function can be defined [44] as in Eq. (2). (2) where: • E refers to the sum of squared differences between the original and moved window.  • u refers to the window's displacement in the x-direction.
• v refers to the window's displacement in the y-direction.
• w (x, y) refers to the weighting function of the window at position (x, y), either a gaussian or a window of ones.
• I (x + u, y + v) refers to the intensity of the moved window.
• I (x, y) refers to the intensity of the original window. Eq. (2) can be further expanded using Taylor's series and rewritten as in Eq. (3).
Eq. (3) can also be rewritten in a matrix form as in Eq. (4).
where A represents Harris-Matrix and is defined as in Eq. (6).
Depending on the value of R, a point is deemed to be a [46]: Shi-Tomasi algorithm detects a corner using two eigenvalues as the Harris-Stephen algorithm does, but it calculates R function differently. In the Shi-Tomasi algorithm, R is calculated as follows: If R is greater than a predetermined threshold, then the selected point is regarded as a corner point [39].
Shi-Tomasi algorithm produces feature points that are more stable and accurate for tracking than the Harris-Stephens algorithm. On the contrary, it results in higher computational demands [42].

IV. THE PROPOSED METHOD
This section describes the proposed video steganography method, which is based on the corner points regions of Shi-Tomasi algorithm and 4-LSBs algorithm. Shi-Tomasi algorithm with a predetermined threshold is used to detect corner points regions (i.e., Region of Interests (ROIs)) in Y (luminance) channel of each frame within the cover video. Whereas, 4-LSBs algorithm is used to embed the secret message in each pixel of the detected regions. To increase the security level of the proposed method, Arnold's cat map algorithm is used to encrypt the secret message before the embedding process. Here, the proposed method uses a square image with RGB color space as a secret message.
Before implementing the proposed method, both the sender and receiver should, in advance, agree upon the cover video, the shared secret key, the image size, and the threshold value. Where the cover video is a video in which the secret message is hidden in it, the shared secret key represents the key used in Arnold's cat map algorithm, the image size represents the size of the secret image to be embedded/extracted, the threshold value represents the value used in Shi-Tomasi corner detector algorithm to detect corner points in the frames of the cover video where different threshold values result in different corner points. Fig. 4 gives a generic design of the proposed method. Message encryption, message embedding, message extraction, and message decryption for the proposed method are explained in the following subsections. Fig. 5 demonstrates the block diagram of message encryption and message embedding for the proposed method. In this stage, the cover video, the threshold value, the shared secret key, the secret image, and the image size are the inputs, and the stego-video is the output. The image size must be square and set by the user due to the use of Arnold's cat map algorithm that only accepts a square image as input. Before the embedding process, the secret image is encrypted with the shared secret key using Arnold's cat map encryption algorithm as described in section III-A. Here, the shared secret key indicates the number of iterations used to scramble the secret image. After that, the encrypted image is partitioned into three images representing Red, Green, and Blue color channels, respectively. Each of these images is then transformed into binary bits. Next, the binary bits of the three images are concatenated to form a single sequence of binary bits that represents the encrypted image. To embed the binary bits obtained from the encryption process inside the cover video, first, the cover video is partitioned into frames. Each Frame is then partitioned into Y (luminance), U (Cb blue chrominance), V (Cr red chrominance) channels. After that, Shi-Tomasi algorithm with a predetermined threshold is applied to the Y channel of each frame in the subsequent processes to detect corner points regions (i.e., ROIs) in them as described in section III-B. Next, the binary bits are embedded in each pixel of the detected regions using 4-LSBs algorithm. Then, the Y, U, V channels of each frame are transformed back into their original color space. Finally, the frames are transformed into a video known as a stego video. Fig. 6 demonstrates the block diagram of message extraction and message decryption for the proposed method. In this stage, the cover video, the stego video, the threshold value, the shared secret key, and the image size are the inputs, and the secret image is the output. Here, Shi-Tomasi algorithm cannot be applied to the Y channel of each frame within the stego video due to the change in pixels values during the embedding process, which will result in different corner points regions (i.e., ROIs). Therefore, the cover video is needed to identify locations of the pixels where the binary bits are hidden in them. To extract the embedded binary bits from the stego video, the cover video, first, is partitioned into frames. Each Frame is then partitioned into Y (luminance), U (Cb blue chrominance), V (Cr red chrominance) channels. After that, Shi-Tomasi algorithm with a predetermined threshold is applied to the Y channel of each frame in the subsequent processes to detect corner points regions (i.e., ROIs) in them as described in section III-B. Then, the embedded binary bits are extracted from pixels of the Y channel of each frame within the stego video in the subsequent processes, so that their locations are equal to the detected locations in the previous step. Next, the extraction process terminates when the length of the obtained binary bits is equal to the length of the (image size × 3) in binary bits. Here ''× 3'' is used because the secret image contains three channels with the same size (red, green, and blue, respectively). Afterward, the extracted binary bits are transformed into three images of size n × n. The three images are then transformed into one image with RGB color space. The image obtained from the extraction process is then decrypted with the shared secret key using Arnold's cat map decryption algorithm as described in section III-A. Here, the number of iterations to be followed to decrypt the image is equal to Arnold's period minus the shared secret key. Where Arnold's period represents the number of considered iterations in a given square image. The image obtained from the decryption process is known as a secret image.

B. MESSAGE EXTRACTION AND DECRYPTION
The following is a numerical example that elaborates of how our proposed algorithm works in detail. Assume the secret message is a color image of dimension 5 by 5, and the cover video data consists of 30 frames per second where each frame has a resolution of 7 by 5. Assume, the corner points detector has extracted two feature points in the Y component of the first frame. The embedding process will take place as the following: Red channel values of the secret image. 32  8  15  67  155  1  61 194 109 249  20  97 64  164 151  211 91 130 66  180  198 11 180 177 63 Firstly, the red channel pixels' values will be converted from decimal (P 1 = 32, P 2 = 8, P 3 = 15, P 4 = 67, P 5 = 155) 10 to 8-bit binary values (P 1 = 00100000, P 2 = 00001000, P 3 = 00001111, P 4 = 01000011, P 5 = 10011011) 2 .
In this assumption the first frame has two feature points, the first feature point is  Embedding process in each video frame can be done by replacing 4-LSBs of each feature point FP with 4-bits of each pixel in the secret image. The following is the embedding process of two feature points in the Y component of the first frame. At the beginning, the first part of the secret message (P 1 ) in the red color P 1 (32) 10 = (00100000) 2 will be embedded into the first feature point. As a result, the first stego feature point value (SFP 1 ) will change from FP 1 (197) = 11000101 to SFP 1 (192) 10 = (1100 0000) 2 . Next, the second part of the secret message (P 1 ) in the red color P 1 (32) 10 = (00100000) 2 will be embedded into the second feature point FP 2 (43) = 00101011. The second feature point after embedding process will be SFP 2 (34) 10 = (0010 0010) 2 as shown below:

A. DATASET
A dataset containing 15 commonly used video sequences, in the 4:2:0 YUV format, was used to evaluate the proposed video steganography method. This dataset was obtained from reference [47]. Fig. 7 shows the first frame for each cover video used in this work. Besides, Table 1 gives a detailed description of all these cover videos. The secret message was chosen to be an image of the logo of the University of Zakho of size 318 by 318 as shown in the Fig. 8. Our work was implemented using MATLAB (R2017b) software program on a personal computer with the following specifications: Windows 10 Pro 64-bit operating system, Intel Core i7 2 nd Generation (8 CPUs) 2.2GHz, Random Access Memory (RAM): 6144MB DDR3, Video RAM (VRAM): Radeon 6000 series 2034MB.

B. EVALUATION METRICS
The challenge of improving/innovating any video steganography methods is to embed as much information as possible in the cover video with a minimum noticeable difference in the stego video. Thus, the proposed method was evaluated and compared with state-of-the-art approaches using two metrics, namely embedding capacity and imperceptibility. Embedding  capacity refers to the maximum amount of information that can be hidden inside the cover video [27], [48], which is measured in bits-per-pixel (bpp) and calculated as in Eq. (9).

Embedding Capacity
= Numberof embedded bits Cover video size in pixels × 100% (bpp) (9) The second metric, imperceptibility, is evaluated by measuring the visual quality of the stego-videos. Usually, to measure this metric, the Peak Signal-to-Noise Ratio (PSNR) is used, which is measured in decibels (dB) and calculated as in Eq. (10). PSNR values falling below 30 dB indicate that the human eye can notice the distortion. Hence, a good steganography algorithm should seek for 40 dB or more [29], [49].
Here, MSE refers to the mean squared error and is calculated as in Eq. (11).
where A and B represent the original and stego frames, respectively, a and b represent the resolution of the given video, c VOLUME 8, 2020  refers to the number of channels that exist in the given color space (for RGB color space, c = 3). MAX A represents the highest pixel value in frame A. Finally, to evaluate the performance of the proposed method against different types of attacks (such as Gaussian noise, Speckle noise, and Salt & Pepper noise), the robustness metric was used. This metric computes the similarity ratio between the original message and the extracted message. Here, the structural similarity index (SSIM) function was used to measure this metric, which is expressed mathemat-ically as in Eq. (12). Higher similarity values indicate better quality of the extracted image [50].
where O represents the original image, E represents the extracted image, µ O and σ O represent the mean and standard deviation values of pixels in image O, respectively, µ E and σ E represent the mean and standard deviation values of pixels in image E, respectively, C 1 and C 2 refer to a fixed value, σ OE represents the covariance between O and E images.

C. RESULTS OF THE PROPOSED METHOD
This section shows the performance of the proposed method in terms of embedding capacity, PSNR, and SSIM on 15 cover videos. Here, before the embedding process, we detected the corner points of each cover video using Shi-Tomasi algorithm with a threshold of 0.00005 to embed the secret data into them. The reason for selecting the threshold of 0.00005 was based on empirical tests in which we could detect more corner points in each cover video without affecting the quality of the stego video. Moreover, we used 4-LSBs method for embedding the secret information within the cover videos. Table 2 shows the number of corner points detected in each cover video. From Table 2, it is clear that the number of detected corner points varies from one cover video to another. As it is also evident from Table 2 that both Paris and Suzie cover videos obtained the highest and the lowest number of detected corner points, which are 1511612 and 67797, respectively. This is due to the difference in frames' scenes and the number of frames in each cover video.
The performance of the proposed method in terms of embedding capacity and PSNR on 15 cover videos has been summed up in Table 3. From Table 3, it can be seen that the videos ''Waterfall'', ''Mother and Daughter'', ''Coastguard'', ''Hall Monitor'', ''Silent'', ''Container'', ''Foreman'', and ''Suzie'' provide a higher embedding capacity rate than other videos. This is because the corner points in these cover videos are abundant. The embedding capacity obtained for the low-corner points videos ''Tempete'', ''Akiyo'', ''Mobile'', ''Flower'', ''Carphone'', ''Paris'', and ''News'' is not as good as the embedding capacity obtained for other videos. This is expected due to the lack of corner points found in these cover videos, where the area of the extracted ROI is very little. It is also evident from Table 3 that the average of embedding capacity in all videos used is 0.069. This indicates that the proposed method can hide an acceptable amount of information. Moreover, as shown in Table 3, the PSNR values of all videos used are greater than or equal to 53.196 dB. This shows the high perceptual invisibility of the proposed method. It can also be noted from Table 3 that the average value of PSNR in all videos used is 61.440 dB. This confirms that the proposed method is highly imperceptible. Accordingly, we can deduce that the proposed method provides a high degree of imperceptibility with an acceptable embedding capacity. Fig. 9 illustrates the performance analysis of the proposed method in terms of the total number of detected corner points and the total number of the embedded bits in each tested video. It is obvious from Fig. 9 that the embedding capacity of each cover video increases whenever the detected corner points increase. Fig. 10 demonstrates the performance of the proposed method in terms of SSIM rate with and without noises on 15 cover videos. From Fig. 10, it can be noticed that the SSIM rate is almost ''1'' in all cover videos when no noise is added to them. This indicates that the secret data can be recovered by losing a little bit of data. It can also be seen from Fig. 10 that when Salt and Pepper noise with a density of 0.002 is added to the cover videos, the proposed method can still provide a high SSIM rate, which is very close to the SSIM rate of cover videos without noises. However, the SSIM rate decreases when the density of Salt and Pepper noise increases. Moreover, from Fig. 10, It can also be noted that when Gaussian noise is incorporated into the cover videos, the proposed method obtains the SSIM rate of less than 0.78 (Flower video), which is still acceptable. But, when Speckle noise is incorporated into the cover videos, the proposed method obtains the SSIM rate of less than 0.71 (News video), which degrades the performance of the proposed method. Fig. 11 shows the extracted secret images from five cover videos (Akiyo, Carphone, Coastguard, Container, and Flower) after different attacks have been applied on them.
We can conclude from Fig. 10 that the proposed method is robust when the cover videos are free of noises, however, the proposed method degrades when noises are incorporated into the cover videos. Besides, the proposed method performs better with Salt and Pepper than other types of noises. VOLUME 8, 2020   In addition, the proposed method with Gaussian noise obtains a higher SSIM rate than with Speckle noise. This is due to the noise nature in changing pixels values of the frames, where the pixels values of the extracted ROI get changed.

D. COMPARISONS WITH OTHER APPROACHES
In this section, the perceptual invisibility and the embedding capacity of the proposed method were compared with the proposed methods in the literature. The methods presented in [27]- [31] were chosen for comparison with the proposed method in terms of PSNR rate. To make the comparison fair, we used the same videos as those used in the methods presented in [27]- [31]. Tables 4 (Part I and Part II) list the PSNR rates obtained by the methods presented in [27]- [30] and the proposed method. The average results are bold-faced. From Tables 4 (Part I and Part II), it is evident that the proposed method attains the highest PSNR rate compared to the methods presented in [27]- [30] across all videos used. It is also clear from Table 3 that the average values of PSNR rate obtained by the proposed method is much better than the methods presented in [27]- [30]. This is about 8.556, 25.623, 26.132, and 23.578 dBs better than the average values of PSNR rate presented in [27]- [30], respectively. To further validate the efficacy of the proposed method in terms of visual imperceptibility, the proposed method was compared with the method presented in [31]. As shown in Table 4 (Part III), the proposed method outperforms the method presented in [31] in terms of average value of PSNR rate. Although the method presented in [31] is better than the proposed method when ''Suzie'' video is used but the differences between them is too little, which is 1.2643 dB.
It is worthy to mention that the proposed algorithm has a linear time complexity O(N) which will grow in direct proportion to the size of the secret data.
The methods presented in [27], [28], [31] were chosen for comparison with the proposed method in terms of the total number of embedded bits. Table 5 (Part I and Part II) list the total number of embedded bits obtained by the methods presented in [27], [28], [31] and the proposed method. The average results are bold-faced. From Tables 5 (Part I), it can be seen that the proposed method attains the highest total number of embedded bits compared to the methods presented in [27], [28] across all videos used. It can also be noticed from Tables 5 (Part I) that the average value of the total number of embedded bits obtained by the proposed method is much higher than those obtained by the methods presented in [27], [28]. This is about 21 and 87 times better than the average values of the total number of embedded bits presented in [27] and [28], respectively. To further validate the efficiency of the proposed method in terms of the total number of embedded bits, the proposed method was compared with the method presented in [31]. As shown in Table 5 (Part II), the method presented in [31] outperforms the proposed method in terms of average value of the total number of embedded bits. Although the method presented in [31] is better than the proposed method but the proposed method is still better in terms of visual imperceptibility.
Although the embedding capacity and visual quality are contradictions, the proposed algorithm have made an excellent balance between both factors. From the results obtained, we can deduce that the proposed method attains an acceptable embedding capacity and obtains better visual imperceptibility than those methods presented in [27]- [31].

VI. CONCLUSIONS AND FUTURE DIRECTIONS
This paper proposes a video steganography method based on the corner points regions and Arnold's cat map algorithm. The proposed method encrypts the secret information using Arnold's cat map algorithm prior to the embedding process to improve the security of the secret information. For hiding the encrypted secret data, the proposed method first detects the ROI in frames of the cover video using Shi-Tomasi corner detector algorithm with a predefined threshold. After that, the proposed method hides the encrypted secret information into the ROI using 4-LSBs algorithm. From the experimental results, it is clear that the proposed method outperforms state-of-the-art methods presented in [27]- [31] in terms of visual imperceptibility. In addition, it can be seen from the results that the proposed method performs better than the methods presented in [27], [28] in terms of embedding capacity. Although the method presented in [31] outperforms the proposed method in terms of embedding capacity, the proposed method is still better in terms of visual imperceptibility. Furthermore, the given results showed the acceptable range with regard to robustness against artificial noises (Gaussian noises, Speckle noises and Salt and Pepper noises with a density of (0.02, and 0.002)). For future works, different corner point detector algorithms can be investigated, and the level of security can be increased by using one of the available publickey cryptography instead of Arnold's cat map algorithm.