A Novel Sequencing Framework for Multi-object Scene Interpretation in Coverless Image Steganography Using DETR and Graph R-CNN

ABSTRACT Image steganography often faces the issue of noticeable embedding patterns, which render conventional LSB‐based techniques vulnerable to both statistical and machine‐learning steganalysis. The present research offers a solution to this problem by presenting a Secure Random Pixel Distribution (SRPD) framework that integrates AES‐based cryptographic randomization along with a lightweight module LSB approach to make pixel changes unpredictable and with minimal distortion at the same time. The technique offers the option of encrypted coordinates, which can be used for either embedding or external deployment, providing flexibility for various deployment situations. Experimental testing on ten benchmark images reveals the excellent performance of SRPD, with PSNR values ranging from 70 to 85 dB, SSIM scores exceeding 0.998, and MSE values as low as 0.0002. The security check through chi‐square and RS analysis provides p ‐values greater than 0.85 and RS deviations less than 0.03%, demonstrating that SRPD is still “NOT DETECTED” by conventional steganalytic methods. Furthermore, the method's linear computational complexity is accompanied by a very low overhead from AES encryption. In summary, SRPD is a steganographic system that is secure, imperceptible, and cost‐effective in terms of computational resources; thus, it can be used for secret communication and privacy‐preserving multimedia applications.

Paper] Combining Pre-trained and Self-supervised Reconstruction for Coded Light-Field Imaging

StarINN: An efficient invertible neural network for image steganography

NayahStego: A Novel Digital Image Steganography Scheme for High-capacity and Secure Data Hiding in Spatial Domain

A novel data encoding method that integrates Variational Mode Decomposition with a new lightweight image coding technique is proposed and applied to the parametric fault diagnosis of DC-DC converters. This approach addresses the limitations of existing fault diagnosis methods, specifically the low diagnostic accuracy and poor noise resistance resulting from inadequate feature extraction. Initially, the parameter fault data of the DC-DC converter is decomposed into multiple modal components using the VMD method. Subsequently, these modal components are processed through Parameter-Weighted Trigonometric Difference coding to emphasize fault features. The enhanced features are then transformed into grayscale images, converting them into two-dimensional datasets for training the diagnostic model. Experimental results demonstrate that the proposed VMD-PWTDIC method achieves a diagnostic accuracy of 99.37%, which is 58.67% higher than four other comparative methods. Furthermore, compared to other methods, the proposed method reduces processing time by an average of 10.25 seconds.

Applications requiring exact image reconstruction always need lossless image coding. A novel method of block based lossless image coding based on the use of the Prediction by Partial Matching (PPM) algorithm combined with two channel coding and adaptive Huffman coding is presented in this paper. In this work, images are segmented into non overlapping blocks, and pixels are efficiently predicted through the application of context modeling using PPM. To improve coding efficiency, a Two-channel coding is employed to separate bit and data streams. The encoded streams are further compressed by a Huffman coding scheme, adaptively adjusting symbol probabilities to local data statistics. The experimental results demonstrate an improvement in the compression ratio while maintaining image quality. Working with statistical and predictive models, the integration of PPM and two-channel and adaptive Huffman coding has created a flexible and robust coding framework. Finally, the proposed method is compared with previous state-of-the-art lossless coding techniques and evaluated in terms of compression efficiency as well as computational behavior, and found to be superior in both aspects. The proposed method demonstrates an average improvement of 46.97% in CR, 32.62% in BPP, and 2.05% in entropy compared to the TIFF, BMP, and LZW methods. This has shown a bright technology in high-fidelity image storage and transmission devices.

Image steganography is a security technique that conceals secret information within digital images in such a way that makes the hidden content imperceptible to human vision and difficult to detect statistically. The main challenge in image steganography lies in achieving an optimal balance among imperceptibility, embedding capacity, and security. To address these limitations, this paper proposes a content-adaptive Least Significant Bit (LSB) steganography framework that integrates saliency-guided embedding, Ant Colony Optimization (ACO)-based dispersion, and hybrid encryption to improve both invisibility and confidentiality. The system embeds secret data in low-sensitivity regions identified by a robust saliency fusion map, minimizing visual distortion. A block-wise ACO mechanism distributes embedding indices spatially across the image to prevent clustering artifacts and enhance undetectability. The framework removes side-information dependency by regenerating embedding indices deterministically from RSA-derived seeds, ensuring reproducibility during extraction. A hybrid cryptographic module combining Advanced Encryption Standard-Galois/Counter Mode (AES-GCM) and Advanced Encryption Standard-Optimal Asymmetric Encryption Padding (RSA-OAEP), with optional Hamming (7,4) error correction, guarantees confidentiality and reliable key recovery. The proposed framework demonstrates near-lossless imperceptibility, achieving PSNR values of 59.7-60.2 dB for 64 × 64 secret images and up to 64.5 dB for 32 × 32 secrets, with SSIM consistently above 0.999 and MSE below 0.07. Under capacity variations with Bits Per Pixel (BPP), the proposed system exhibits a clear rate-distortion behavior as PSNR decreases from 61.23 to 55.17 dB, while SSIM remains above 0.9978. All index-selection modes (ACO, random, perm, and saliency_topk) differ by less than 0.5 dB, confirming the stability of the LSB-robust content map, with saliency_topk achieving the highest PSNR ≈ 60.16 dB. From a security perspective, the proposed ACO-guided scheme exhibits random-level detectability against modern CNN-based steganalyzers. Overall, the proposed method offers high imperceptibility, stable and lossless extraction under no-attack conditions, accurate payload rate control, and strong resistance to CNN-based steganalysis.

In the rapidly evolving field of digital security, this study aims to advance image steganography by developing and benchmarking seven deep learning architectures with a focus on imperceptibility, embedding capacity, and robustness against steganalysis. The models implemented include the residual dense network (RDN), vision transformer with adaptive attention (ViT-AA), progressive generation network (PGN), dual-stream architecture (DSA), wavelet-based hybrid network (WHN), mutual attention transformer (MAT), and efficient attention pyramid transformer (EAPT). Using the PyTorch framework and standardized datasets such as DIV2K, COCO, and ImageNet, each architecture was trained through structured preprocessing and evaluated using metrics including PSNR, SSIM, LPIPS, and statistical steganalysis resistance. Experimental results demonstrate that WHN achieved the highest visual quality (PSNR = 43.5 dB, SSIM = 0.995), while MAT and EAPT provided superior security with detection rates near random chance (0.501–0.502) and robustness against JPEG compression and noise insertion. PGN and DSA offered low-latency performance suitable for resource-constrained or mobile applications, while ViT-AA provided a balanced trade-off across imperceptibility and robustness. The findings confirm that deep learning approaches surpass traditional methods and establish new computational benchmarks for covert communication and digital forensics. These results recommend WHN, MAT, and EAPT for high-security contexts, PGN and DSA for embedded platforms, and ViT-AA as a general-purpose framework, while encouraging further research into lightweight variants for IoT and real-time deployments. Tools for data collection and experimentation included benchmark datasets and PyTorch-based implementations.

Hog based high capacity and blind image steganography technique with tamper detection and content authentication

This study introduces a novel approach that integrates Gaussian Mixture Models (GMM) with MD5 hash-based verification to detect hidden messages embedded via Least Significant Bit (LSB) steganography in JPEG images. Unlike previous methods, the proposed dual-layer technique combines probabilistic modeling with data integrity verification. The model was trained and evaluated using a dataset comprising both original and stego-JPEG images. The experimental results achieved an accuracy of 78.67% and a precision of 89.15%, indicating good class separation between stego and non-stego images (AUC-ROC = 0.8659). However, the recall rate of 69.70% suggests that there is room for improvement in detecting all stego instances. Although MD5 is a hash function rather than an encryption algorithm, it effectively aids in identifying data anomalies resulting from message embedding. Overall, this lightweight approach offers a practical solution for steganalysis and can be further enhanced through the integration of hybrid deep learning techniques in future research.

Quantifying stellar parameters and magnetic activity for cool stars in double-lined spectroscopic binaries (SB2s) is not straightforward, as both stars contribute to the observed composite spectra and are likely variable. Disentangled component spectra allow a detailed analysis of the magnetic activity of the primary component. We aim to separate the spectra of the two stellar components of the HR,7275 SB2 system. We also aim to obtain a more accurate orbital solution by correcting the observed radial velocities (RV) from activity perturbations of the spotted primary (``RV jitter'') and to derive a surface image of this component. We obtained time-series high- and ultra-high resolution optical spectra and applied two different disentangling methods. We modeled radial velocity (RV) residuals using three-sine function fits and modeled the spectral-line profile of the primary with the Doppler imaging code iMAP. We measured magnetic fields for the primary based on least-squares deconvolved Stokes-V line profiles, and determined chromospheric emission from the line cores of ̧ahk, ̧airt,8542,Å, and Balmer ̋alpha. We first applied our disentangling technique, which allowed us to determine the properties of the system more accurately before performing these analyses. The Doppler image of the primary shows two large cool spots that cover approximately 20% of the visible hemisphere, plus three smaller spots each still covering approximately 13% in size. Between May-June 2022, HR,7275a exhibited an impressive spottedness of roughly 40% of its entire surface. The RV is modulated by the rotation of the primary, with maximum amplitudes of 320, for HR,7275b, which is an extremely low value. and 650, _ for two different modulation behaviors during the 250,d of our observations. This jitter is primarily caused by the varying asymmetries of the apparent disk brightness due to the cool spots. Its removal results in roughly ten times higher precision of the orbital elements. Our snapshot magnetic-field measurements reveal phase-dependent (large-scale) surface fields between +0.6±2.0,G at phase 0.1 and -15.2±2.7,G at phase 0.6, indicating a complex magnetic morphology related to the location of the photospheric spots. We also obtain a logarithmic lithium abundance of 0.58±0.1 for HR,7275a, indicating considerable mixing, and 0.16 +0.23 -0.63

Abstract The rapid growth of digital communication has increased the demand for secure and tamper-proof information exchange. This project presents CryptoSteg Secure Suite, a hybrid security platform that integrates AES-based symmetric encryption with LSB image steganography to achieve confidential and covert message transmission. By combining cryptography with data-hiding techniques, the system delivers a robust multilayered security model capable of protecting sensitive data against modern cyber threats. In this approach, sensitive messages are first encrypted using the Advanced Encryption Standard (AES), a fast and industry-accepted symmetric encryption algorithm that ensures strong confidentiality with efficient processing. The encrypted data is then embedded into digital images using the Least Significant Bit (LSB) steganography technique, which hides information within pixel values without causing noticeable visual distortion. This ensures that both the content and the existence of the message remain protected. The system is implemented with a user-friendly Python Tkinter graphical interface, allowing users to easily perform encryption, embedding, extraction, and decryption operations. Additional features such as input validation, error handling, and image-quality preservation enhance reliability and usability. Experimental results demonstrate that the generated stego-images maintain high visual quality while securely carrying encrypted data. By integrating AES encryption with LSB steganography, CryptoSteg Secure Suite demonstrates a practical and effective secure communication model. While AES protects the message content, steganography provides stealth by concealing the presence of data itself, for future enhancements, offering a scalable foundation for research in secure communication, digital forensics, and privacy-preserving technologies. Key words: AES Encryption, Symmetric Cryptography, LSB Steganography , Secure Communication Python Tkinter.

By eliminating the need to alter the source images, this paper introduces a secure technique for coverless image steganography that strengthens defense against steganalysis attacks. Our method makes use of a hybrid Generative Adversarial Network (GAN) with a Support Vector Machine (SVM), which is trained and validated on a Diffusion Weighted Imaging (DWI) dataset to retain visually indistinguishable steganographic representations while increasing security. A powerful feature extraction capability of several Deep Learning Models (DLMs), EfficientNet-B4, DenseNet121, and Residual Network-18 (ResNet-18), integrated with the Vision Transformer (ViT) is performed. With the highest Peak Signal-to-Noise Ratio (PSNR) of 45.87 dB and Structural Similarity Index (SSIM) of 0.98, the ViT-GAN-SVM model exceeds other suggested models in terms of steganographic quality. Additionally, the ViT-GAN-SVM system achieves 99.78% accuracy, 99.85% sensitivity, 98.99% precision, and 99.85% F1-Score in terms of diagnostic accuracy. The ViT-GAN-SVM model performs much better than other introduced models in all diagnostic performance metrics, with increases ranging from 5.55% to 6.36%. This shows that ViT-GAN-SVM is a superior choice for medical diagnostic tasks since it can correctly identify prostate cancer on the DWI prostate cancer dataset.

Abstract To achieve accurate recognition of objects grasped by a three-fingered flexible bionic robotic gripper, this study proposes an identification method based on Fiber Bragg Grating (FBG) sensing and Markov image coding. Firstly, the grasping process of the manipulator was analyzed through ANSYS finite element simulation. The stress and strain distribution on the inner side of the gripper was simulated under three load conditions (3N, 5N, and 7N) to determine the optimal packaging position of the FBG sensor. Subsequently, FBG sensors were packaged on the inner sides of the three fingers of the manipulator, and grasping experiments were conducted on nine objects with different diameters, materials, and shapes. For the three-channel sensing data, a Markov Transition Field (MTF) was used for fusion coding to convert time-series tactile signals into image features, and a strategy of using self-defined weights was proposed to construct the image dataset. Finally, the VGG11 convolutional neural network was employed for classification training and testing of the three datasets. Experimental results show that: compared with other weight coding methods, the self-designed weight method achieves an accuracy of 97.14% with an average recognition time of approximately 0.31s per object. Moreover, the self-designed weight coding performs better in terms of loss and accuracy convergence on training and validation sets, with stable loss free of fluctuations, sustained increase in training accuracy, stable generalization, and efficient fitting. This study verifies the effectiveness and reliability of the proposed method in the recognition of objects grasped by the manipulator, and provides a feasible solution for the tactile perception and object recognition of flexible bionic manipulators.

Digital image steganography is the art and science of hiding secret information in an innocent looking cover image to covertly exchange sensitive information in real-world scenarios. This paper presents a transform-domain steganographic method that leverages the Discrete Wavelet Transform (DWT) and a skin-based masking mechanism to identify perceptually less sensitive regions for embedding while maintaining high imperceptibility and extraction accuracy. The proposed method extends our previous work using S-transform which is an integer-to-integer discrete wavelet transform (DWT). The hiding process starts with dividing the cover image into the basic color channels and applying DWT on each channel independently. The approximation coefficients of the DWT are then used to build a blocked skin-map. Only a pixel marked as “skin” in the blocked map will cause its corresponding approximation coefficients to be embedded with the bits of the secret message. Experimental results demonstrate that the proposed approach achieves competitive performance in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), outperforming several existing methods. Limitations and future directions, including robustness to geometric distortions and steganalysis detection, are discussed.

Today, communication using digital media has increased rapidly. In this digital communication era, providing security to sensitive images is essential at the time of transmission. The images are mostly focused nowadays because they are used directly or indirectly in every field of information sharing, for example, healthcare, military, intellectual property, and many more areas. In this paper a new approach to image scrambling is proposed to secure the sensitive image information. The proposed method is focused on the core concept of data structure. It involves the use of binary trees and the efficiency of hash tables, which enhances image security during transmission. The dynamic data structure properties enhanced the scrambling and descrambling process. In the scrambling method, the image pixels are first stored in the binary tree using the hash table. After the binary tree arrangement, pixels are collected into a one-dimensional array using the tree traversing process. Now, the one-dimensional array is converted into the two-dimensional array to match the size of the original image. The descrambling method is the inverse of the scrambling method. The proposed method maintains the quality of the image for both sender and receiver; different quality assessment parameters like PSNR, MSE, NCC, AD, SC, MD, Corr, HC, VC, DC, NPCR, UACI and Entropy are used to check the outcome. The outcome of PSNR between the original image and the scrambled image is less than 4 dB. For the descrambled image and the original image, the PSNR is infinite. According to the obtained results, there is a 100% similarity between the original image and the descrambled image. The proposed method was also compared with the existing methods, and it showed a negative or near to '0' correlation between the scrambled image and the original image. In future work the proposed scrambling method can be used in image watermarking or image steganography techniques.

Abstract There are many cover selection methods currently in use within steganography to ensure that secret data embedded in digital images is hard to detect. This paper proposes an improved steganalytic method to enhance the detection accuracy on steganography in digital images by considering image texture complexity. We propose a measurement of image texture complexity based on gray level co-occurrence matrix-based (GLCM) multi-scale fusion, which is integrated into the classifier’s detection process. Our method refines the traditional classification process by exponentiating the probability of an image being voted “clear” with its calculated complexity value and comparing this value to the probability of it being voted “stego”. Images with higher texture complexity are more likely to be identified as containing secret data. Experimental results show that our method can effectively increase detection accuracy of steganalysis when cover selection is employed in steganography.

The exponential growth of internet usage for transmitting sensitive information has intensified the demand for advanced security techniques to safeguard digital communication. This study aimed to enhance image steganography by modifying the Least Significant Bit (LSB) method with the J5 algorithm and applying file compression to improve data security and reduce detectability. The method involved embedding hidden text into digital images using a C#-based J5 steganographic tool, secured with password-protected extraction, and evaluating performance with machine learning techniques such as accuracy scoring, confusion matrices, and structural similarity analysis. Results showed that while embedding messages increased pixel modifications and reduced accuracy with larger payloads, the integration of compression reduced inflated stego-image sizes by over 85%, thereby minimizing suspicion without loss of hidden information. Comparative analyses demonstrated that the proposed approach achieved peak signal-to-noise ratio (PSNR) values that were comparable to, and in several cases higher than, those obtained in existing studies, while consistently yielding lower mean squared error (MSE) values across the tested image datasets. In conclusion, this work validates the feasibility of combining steganography, machine learning, and compression to achieve more practical, secure, and efficient data hiding in modern communication systems.

Image steganography is the process of hiding information, which can be text, image, or video inside a cover image. Recent steganography literature hasn't addressed the problem of loss of secret information during extraction and reliability. Hence, to reduce information loss and provide reliability between in the basic criteria, Herein, we proposed a hybrid method that utilizes the least significant bit (LSB) substitution, transppsition, magic matrix, key and Advance Encrytion Standard (AES) algorithm. The LSB method decreases embedding errors by implementing a new value difference algorithm. In addition, to improves the reliability between the basic criterion for image steganography we used transposition, magic matrix, key and AES. The proposed method ensures a high-quality image format in the RGB color model to conceal the hidden message within the cover image which is jpeg. The proposed hybrid method performed several experiments and these are mainly based on quality assessment metrics such as PSNR, SSIM, RMSE, NCC, etc. which showed better results. The proposed method also analyzed with different perspectives in terms of different dimensions of images and different sizes of message text which showed better results. In addition, the performance of the proposed method showed better results based on (regular and singular) steganalysis, noise, and cropping attacks. The security analyses such as key space, differential, and statistical attacks show that the proposed scheme is secure and robust against channel noise and JPEG compression.