Enhancing data security using Laplacian of Gaussian and Chacha20 encryption algorithm

Abbreviations

LoG

Laplacian of Gaussian

PSNR

peak signal-to-noise ratio

MSE

mean square error

NPCR

number of pixels change rate

UACI

unified average changing intensity

Exp

exponential

Nomenclatures

G

Gaussian kernel

X

X coordinate

Y

Y coordinate

Δ

Delta

Greek symbols

σ

Standard deviation

∂

Derivative

π

Semi-vertex angle of the conical nose (Figure 1), rad.

1 Introduction

Due to the continuous evolution of information security, protecting sensitive information is considered ultimate requirement in various applications. Using hybrid techniques to conceal encrypted data within multimedia environment is considered a powerful method to perform secure communication [1]. This study suggests to integrate two strong algorithms, Laplacian of Gaussian (LoG) edge detection technique and ChaCha20 encryption algorithm in order to select the appropriate locations for concealing the encrypted data inside video frames.

LoG is a powerful edge detection algorithm that support concealing information within perception visual content, while ChaCha20 encryption algorithm is used to strengthen concealing the encrypted data due to its speed and secrecy [2].

There are several issues and challenges that are addressed by the proposed approach which are discussed as follows. Many of the current edge detection techniques face difficulties in choosing the suitable edge accurately for concealing data within frames especially with complex video scene that may lead to visual distortion. In terms of security, using a weak encryption method or an inadequate key management system may expose sensitive data to different attacks. Further, achieving a balance between data concealment and perceptual impact such that the perceptual effect is not noticeable is difficult, especially with the huge amount of data to be concealed. High complexity represents an important challenge for real-world applications, where time is considered an important factor; it is also a difficulty for limited resource systems.

The proposed system addresses the discussed challenges through several means. Using the LoG algorithm augments the accuracy in selecting the suitable edge location for embedding data that prevents noise with high reliability and reduces the inappropriate selection of the edge location. Further, using the ChaCha20 encryption algorithm supports concealed data security due to its features of speed, high performance, and immunity to different types of well-known attacks. The proposed system guarantees imperceptibility due to the careful selection of the concealing location to reduce virtual distortion by integrating edge detection and encryption.

The LoG method represents the backbone in the proposed system because it offers strong approach to combine encrypted data in a visual media. As a result, the concealed data remain unseen to human eye and keep the video content uncorrupted.

The integration between LoG and ChaCha20 techniques support the strength for both of them, as well as it is a new era for securing sensitive information using video frames. The purpose of Merging these methods together is to keep secret information immune against intruders and attacks and securing protecting information during storage and transmission.

The suggested combination of these methods can be considered a novel method for a secure encryption by concealing the ciphertext within digital video. This hybrid method can be employed in different applications like secure end to end communication, video conference, and surveillance.

The aim of this study is to combine strong algorithms to develop a powerful novel method for embedding encrypted data within video frames, which support resistance against unauthorised access. The proposed approach aims to enhance the accuracy of selecting a suitable location for embedding the encrypted data with the least visual distortion and ensure the security of the encrypted data, by concealing them inside the video frame to increase the difficulty in detecting them. An additional objective is to maintain the quality of visual perception of the video frames so that the embedding process remains unpredictable for intruders. The following are the main contributions of this study:

• Protecting important details and reducing the visual effects due to the concealment of the encrypted text inside the chosen locations of video frames.

• Combining LoG edge detection and ChaCha20 algorithms to increase security. It provides multilevel security, especially, ChaCha20 is a fast and strong defence against many cybersecurity attacks.

• Providing a balance between the security level and efficient visual quality.

Concerning the remaining sections of the study, Section 2 presents related works, Section 3 explains the methodology, Section 4 provides the experimental results, Section 5 discusses comparison analysis, Section 6 discusses limitations and Section 7 presents the conclusion.

2 Related works

This section discusses the previous works that used different edge detection techniques in diverse applications.

One of these is the Edge Adaptive based on the Least-Significant-bit Matched Revisited (EALSBMR) approach [3]. The method detects edges using image processing and develops a comparison using MATLAB software. To identify edges, the Sobel approach is used on cover images. Then, secret bits are embedded by taking advantage of sharper edges. In this research, the EALSBMR method is used with Sobel operator assistance to achieve an application that is both faster and more reliable than the EALSBMR technique, with a reasonable compression ratio. To get edges, cover images are subjected to the Sobel edge detection approach. Subsequently, sharper edges are used to insert hidden information. This method offers a decent compromise between image quality and security. On the other hand, the strength of the method against different types of attacks is not measured accurately.

Two basic edge detection limits are presented in another study: edge thickness and edge connectivity [4]. A reliable edge detection technique using multiple threshold approaches (B-Edge) is suggested. The commonly used canny edge detector observes some gaps in optimal results and concentrates on selecting two thresholds. This approach chooses the simulated triple thresholds that target the key issues with edge detection: picture contrast, choosing effective edge pixels, solving mistakes, and resemblance to the ground truth. This closes the loopholes of the cutting-edge operator. The suggested method aims to improve both coloured and grayscale photos. However, the suitability of the method for different applications in real world is not considered.

Using the multi-criteria decision-making approach and visual cues from satellite pictures, Bausys et al. [5] suggested an approach involving the adaptive selection of edge detection techniques. Selecting the best technique for the selected satellite photos is not easy because no single right algorithm works for all situations. Instead, it depends on a variety of elements, including the raster image’s content and acquisition, real-world image visual characteristics, and human vision. This method suffers from limited exploration in different environments or imaging models.

Xui and Ge [6] suggest a deep learning approach for image smoothing and edge identification, depending on the convolutional neural network. The findings demonstrate that the research strategy suggested in this study greatly enhances the edge effect, preserves the usefulness of edge information, and effectively addresses the issues of detecting edges and graping information. Simultaneously, it enhances the smoothing effect of the image and lowers the signal-to-noise ratio (SNR) of the smoothed image. Evaluating the method’s performance and its robustness vis-a-vis other techniques for complex structured images is not considered in this study.

Xiaofeng et al. [7] suggested a detecting edges approach based on deep learning for cancer images, since the current edge detection for medical images is low in both image restoration accuracy and fitness of the optimisation coefficient, leading to detection outcomes with low information recall, bad smoothness, and weak detection accuracy. First, a three-dimensional model of the cancer image’s surface structure reconstruction is created. Second, the fine-grained characteristics of the cancer cells in the cancer image are extracted using edge contour feature extraction approach. Ultimately, the multi-dimensional pixel feature distributed reconstruction model of the cancer image is built. The ultra-fine particle feature is recovered, and regional fusion and information reconstruction are realised by using the fine-grained feature segmentation approach. A combination deep learning system enables the adaptive optimisation of edge detection. Evaluating the performance of the method for different and complex structure images is not explored in this study.

Academics working on different applications currently find it difficult to identify underwater images, including tracking fish species, monitoring coral reef species, and counting marine life [8]. As underwater images are often subjected to distortion and light attenuation, pre-processing techniques are necessary to improve their value. The edges of the underwater photos were identified in this article through a variety of edge detection methods. A variety of specialised approaches, including thresholding, median filtering, Wiener filtering, and enhancement processing, are used to pre-process the images. For evaluating the effectiveness of each edge detection technique employed in the testing, coral reef photos are used as a dataset of underwater images. An underwater GoPro camera is used to take pictures of every coral reef dataset. However, evaluating the performance of the method in different environments and with different underwater images is not addressed.

In remote sensing processing, edges are discrete geometric features essential for higher-level object detection and recognition [9]. However, because of the significant speckle noise that causes false positives (type I errors), edge detectors intended for optical images typically perform poorly on SAR images. Consequently, numerous researchers have proposed edge detectors specially designed to address the features of SAR images. While these edge detectors may yield good results in independent assessments, the comparisons typically include a rather small set of (simulated) SAR images. Due to this, the generalised performance of the suggested approaches is not accurately reflected because real-world patterns are far more varied and complex. This leads to another issue, which is the absence of a quantitative standard in the industry. Therefore, it is currently impossible to compare fairly any edge recognition technique for SAR images. Therefore, this study aims to fill up such gaps by offering a thorough experimental assessment of edge recognition in SAR images. To achieve this, a benchmark is presented on SAR picture edge recognition techniques, by assessing several publicly available techniques, including those considered the cutting edge. The limited exploration of the method’s performance with noise is a gap.

Arulananth et al. [10] presented the fast Pixel-based matching and contour mapping approach, which uses non-local and mask-propagation approaches for edge comparison of the reference and targeted frames. Since our approach uses information from both the first and previous frames, it can handle obstacles and resist noticeable item’s visual fluctuation. Section 4 discusses the improvement in the performance of the suggested system, with tabulated and drawn data. The most notable reinforcements are found in detection probability and detection time. There is no comparison with other methods or any evaluation with complex images.

Sultana et al. [11] proposed a hybrid image steganography method of secret hiding by combining the edges and local binary pattern (LBP) code qualities. Only edge pixels that affect how successfully a new methodology conceals data are used in this strategy. A logical OR operation is used to apply multiple edge detectors and hybridise them to enhance the number of computed edge pixels. For this, a morphological dilatation process in the hybridised edge picture is used. For edge pixels, least significant bits (LSB) and all LBP codes are computed. These LBP codes, LSBs, and secret bits are then combined using an XOR operation. The LSBs of the edge pixels receive these inserted bits consequently. The limited exploration of the method’s strength against different attacks and its performance in concealing different types of data inside images are not measured.

Sultana et al. [12] used additional edge pixels by employing edge detectors in the prediction error space (PES). The PES is generated by using a predictor on the cover image and then computing the prediction errors. The edge detector is used to mark the edges in PES. More information is acquired by the edge-error corresponding pixels than by the pertinent pixels, which does not result in an edge-error. To generate more edges and a better embedding capacity, the output from many edge detectors is also aggregated. In non-edge pixels, where x > y, x number of hidden bits and y number of bits are implanted. According to the simulation results, the suggested scheme performs better than its competitors on every performance-measuring criterion, such as payload, stego picture quality, and assault resistance. Its scalability for concealing different data applications in diverse images is not explored.

3 Methodology

This section discusses the algorithms employed to design the proposed system.

3.3 Proposed system algorithm

As mentioned, some significant texts may surface in video recordings that are confidential and should not be shown to the general public. A novel algorithm has been proposed that chooses which texts in the video should be encoded. The main purpose of this approach is to encrypt text images extracted straight from the input video frames. A video clip is the input (with hidden texts anticipated). The following diagram (Figure 1) handles the partially encrypted video. The suggested technique is explained by Algorithm 1.

Algorithm (1), Encryption phase
Input: encryption key, video, secret text
Output: encrypted video
Start
Step 1: load video
Step 2: frame extraction
Step 3: convert frames to grayscale
Step 4: apply LoG edge detection
Step 5: select appropriate locations
Step 6: Generate encryption key using ChaCha20
Step 7: XORed text and encryption key
Step 8: embed the encrypted text into the selected locations
Step 9: collapse the frames
Step 10: release the resulting video
End

Figure 1

The proposed system.

The edge detection process involves several steps. First, the image is smoothened using Gaussian blurring to minimise noise and address the greatest intensity changes and then the Laplacian filtering is done to improve the edges by exploring intensity gradient changes. Then, a threshold value is determined to discriminate between noise and edges. The edges with a value greater than the determined threshold are kept for further manipulation, to eliminate weak edges that are not suitable for the concealing process. The characteristics considered for choosing the suitable edge for concealing are strength, length, connectivity, and orientation, to minimise noticeable distortion. Next the region of interest (ROI) is determined within the selected edge, where secret data are to be concealed. ROIs are chosen based on the length and volume of the concealed data. Finally, the chosen edge and ROIs are adjusted to guarantee the best performance with the least visual impact. It may involve selected edge fine-tuning, ROI size, and location adjustment and threshold parameters refinement.

The XOR operation plays a principal role in creating the ciphertext. It is employed between a byte of the plaintext and a byte of the encryption key. The same key is used in the encryption and decryption processes. The importance of the XOR operation is represented by its reversibility, randomness, unpredictability and computational efficiency. All these features support the speed, performance and security of the ChaCha20 algorithm.

The system parameters comprise

1. Gaussian Blur of size 7 × 7,

2. Standard deviation to decide the image blurring degree,

3. Laplacian filter of size 7 × 7,

4. Encryption key of 256-bit length,

5. Nonce of 96-bit size, and

6. The number of encryption rounds is 20.

We recently employed text encryption to encrypt some text, and the outcome was excellent. The suggested method was thoroughly evaluated using various input colour picture samples taken from the input video, and it was shown to be secure and reliable, encrypting data is a wonderful way to protect it and ensure access when needed. The stages involved in encryption are as follows.

3.4 Text detection

The final section of this study explains how the LoG edge detector works and how it was used to find the writings in the movie. Of course, this process involves several mathematical calculations in the background. To explain everything, though, could be a bit intimidating. Consequently, the first three stages are indicated, describing the ideas underlying its functioning.

Sliding a fixed-size window across the image yields the biggest intensity fluctuation while moving in the X and Y axes. For every window found, a score R is determined. The necessary edges are then selected by applying a threshold to this score. For a flat area, R is frequently perceived as being small. A region is considered to be an edge if R is small, and an edge if R is large (Figure 2).

Figure 2

Frame represents book.

3.5 Keystream generation

It is widely recognised in cryptology that the key, the method of generation, and its unpredictability degree are significant variables in deciding how powerful the method is. Cause randomness tests are required for checking the resulting key. It is therefore difficult to hack. The key creation procedure is employed in the ChaCha20 encryption method.

3.5.1 Text encryption

The points are identified for encryption purposes by XORing with the keys after edge detecting, which are represented by the texts for the individual frames (Figure 3).

Figure 3

Detecting text in book frame.

4 Results and discussion

Encryption is the best security technique for protecting digital video footage. Consequently, there are now two ways to encrypt digital video files: completely or partially. The key is essential to the ciphering operation because it enhances the approach, increases security, and makes it harder for intruders to crack. Three video clips were used to determine the measurement results using the suggested algorithm. The ciphering and deciphering results for the three selected samples used in the research findings are explained in Table 1. The suggested encryption methods have shown promising results concerning both time and the encryption process. Table 1 uses three different-sized films to illustrate these values (Figure 4).

Figure 4

Encryption outcome of Figure 3.

Other measurement approaches were used to more accurately and objectively determine the amount of encryption in video encrypting, as human analysis is not the only way therefor. Numerous quality criteria exist. Table 2 displays the most significant ones (unified average changing intensity (UACI), peak signal-to-noise ratio (PSNR), number of pixels change rate (NPCR), mean square error (MSE), correlation, and entropy) used to measure the suggested technique. These criteria are numbered as follows: (1 = MSE, 2 = PSNR, 3 = Correlation, 4 = NPCR, 5 = UACI, and 6 = Entropy).

Table 2 indicates that the proposed method was both effective and resistant to attacks, as evidenced by the high MSE results and low PSNR values found. The recommended algorithm’s correlation values were very good. The algorithm’s strong defence against differential attacks was illustrated by the NPCR and UACI figures. The outcomes of the suggested method are excellent because they are close to the ideal value of entropy, which is known to be 88. This indicates that the recommended approach is successful in distributing the ciphered frame’s pixels randomly. Importantly, the outcomes of the suggested system are good because they are close to the ideal value.

5 Comparison analysis

This section compares the proposed system with the related works mentioned in Section 2, involving several features as described in Table 3.

6 Limitations

The proposed system has several limitations, as indicated below:

• The main dependence is on edge detection and selection of the suitable edges. If the choice is not appropriate then there will be a noticeable distortion that reveals the embedded data

• There could be a variance in the performance of the method depending on the used video type. If the video contains complex contents or there are more changes in the visual elements, it represents a challenge that may reduce the method’s efficiency or increase the visual distortion.

• The limited embedding capacity in cases where the video contains no suitable edges to conceal the encrypted data makes embedding of large encrypted data more difficult and leads to noticeable distortion in the visual contents.

7 Conclusion and future work

This study presents a novel approach for concealing encrypted data inside video frames by combining two strong approaches, LoG edge detection algorithm and ChaCha20 encryption algorithm. This is done by selecting the concealing locations by means of LoG edge detection algorithm and encrypting data using the ChaCha20 encryption algorithm. This combination offers a high level of security without affecting the visual quality. This is due to LoG ability to dedicate locations with high edge information to conceal with. On the other hand, the ChaCha20 encryption algorithm is a lightweight, fast, and strong encryption method which has immunity against many attacks that provide high security for the suggested method. This combination resulted in reducing the distortion in the visual contents and provide strong resistance against many attacks. This study provides a strong and secure method for many applications in the real world such as video surveillance, copy right production, and video conference, i.e. it provides a very secure and efficient method for video communication. Experimental results depending on metrics such as UACI, PSNR, NPCR, MSE, correlation, and entropy, prove the efficiency of this method.

There are several suggestions for future work, including an error correction code with the encrypted text before embedding inside the video, to reduce the impact of noise and error during embedding and transformation. Further, building a strong key management system that involves key generation, exchange, and storage, which involve key rotation and key exchange approaches, will support the encryption method and increase security.