Article [3] discussed how to combine trained CDNs in a multimodal framework, and it examined their detection accuracy. The framework detects each classification modality independently and combines their estimations to create a universal image steganography detector. Six of the latest CDN-based image steganography detection techniques—GNCNN, IGNCNN, XuNet, YeNet, YedroudjNet, and the improved IGNCNN—were trained on stego images generated using WOW, SUNIWARED, and HILL steganography algorithms with payloads of 0.2, 0.3, and 0.4 bits per pixel. Due to the projected similarities between the image steganography systems, the detection accuracy decreased slightly. However, the multimodal image steganography detection based on the improved IGNCNN universal image steganography detection performed best compared to the other five examined detectors [3].

Article [9] discussed detecting steganography-modified JPEG images, and it analyzed image steganography detection using shallow and deep learning methods. Three common stenographic algorithms—JPEG universal wavelet relative distortion (J-Uniward), nsF5, and uniform embedding revisited distortion (UERD) at two density levels—hid information in BOSS database photos. DCTR and GFR were the best feature spaces validated. At 0.4, the nsF5 algorithm detected bpnzac density with 99.9% accuracy, but the J-Uniward algorithm was barely detectable at 0.1 (a maximum of 56.3%). The study concluded that ensemble classifiers were a promising alternative to deep-learning-based detection [9].

In [10], the authors presented a deep-learning-based approach for steganography detection in digital images. The authors began by describing the importance of steganography detection in the field of digital forensics and highlighted the challenges associated with it. They then proposed a CNN-based model for detecting the presence of hidden data in digital images. The proposed model took as input the pixel values of an image and learned to identify the presence of steganography through a series of convolutional, pooling, and fully connected layers. The authors evaluated the performance of the proposed model on a dataset of stego images and showed that it outperformed existing steganography detection techniques in terms of accuracy, precision, and recall. The results of the study suggested that deep-learning-based approaches can be effective for steganography detection in digital images and can help improve the accuracy and reliability of forensic investigations.

In [11], the authors began by describing the importance of steganalysis in digital forensics and highlighted the limitations of traditional steganalysis techniques. They then proposed a deep-learning-based model for steganalysis that used a combination of convolutional and fully connected neural networks. The proposed model was trained and evaluated on a dataset of stego images that contained spatially embedded hidden information. The authors showed that the proposed model outperformed existing steganalysis techniques in terms of accuracy and sensitivity to different types of spatial image steganography. They also performed a sensitivity analysis of the proposed model to evaluate the impacts of different hyper parameters and architecture choices on the model’s performance. The results of the study suggested that deep-learning-based approaches can be highly effective for the steganalysis of spatially embedded hidden information in digital images, and that careful selection of hyper parameters and architecture choices can further improve the performance of a model [11].

The authors of [12] began by describing the importance of steganalysis in digital forensics and highlighted the limitations of traditional steganalysis techniques. They then proposed a deep-learning-based model for steganalysis that used a combination of non-local blocks and multi-channel convolutional networks to identify the presence of hidden information in an image. The proposed model was trained and evaluated on a dataset of stego images that contained spatially embedded hidden information. The authors showed that the proposed model outperformed existing steganalysis techniques in terms of accuracy, precision, and recall. They also showed that the proposed model could be used to localize the regions of an image that contained hidden information. The results of the study suggested that deep-learning-based approaches can be highly effective for the steganalysis of spatially embedded hidden information in digital images and that the proposed model can help improve the accuracy and efficiency of forensic investigations [12].

Article [13] presented a deep-learning-based approach for detecting steganography in color images. The authors began by describing the importance of steganalysis in digital forensics and highlighted the challenges associated with detecting hidden information in color images. They then proposed a multi-frequency residual convolutional neural network (MRF-CNN) for steganalysis that extracted features from different frequency components of an image and learned to identify the presence of hidden information. The proposed model was trained and evaluated on a dataset of stego color images and compared with existing steganalysis techniques. The authors showed that the proposed MRF-CNN model outperformed existing steganalysis techniques in terms of accuracy, precision, and recall. They also showed that the proposed model could be used to localize the regions of an image that contained hidden information. The results of the study suggested that deep-learning-based approaches, such as the proposed MRF-CNN model, can be highly effective for the steg analysis of images and can help improve the accuracy and efficiency of forensic investigations.

The authors of [14] began by describing the importance of hand movement identification in various applications, such as prosthetics and rehabilitation. They then proposed a machine-learning-based approach for identifying hand movements that involved the use of electromyography (EMG) signals recorded from muscles in the arm. The proposed approach used a combination of feature extraction techniques and classification algorithms to identify hand movements. The authors evaluated the proposed approach on a dataset of EMG signals recorded from multiple subjects and showed that the proposed approach achieved high accuracy in identifying hand movements. They also compared the performance of the proposed approach with that of existing approaches and showed that the proposed approach outperformed them. The results of the study suggested that machine-learning-based approaches can be highly effective for identifying hand movements and can have important applications in prosthetics and rehabilitation [14].

In [15], the authors presented a novel deep neural network for identifying contextual steganography methods. The suggested method employed a high-boost filter to reduce high-frequency noise while keeping low-frequency information intact. Thirty high-pass SRM filters were applied to the high-boost image, resulting in thirty high-boost SRM-filtered photos. The suggested CNN used two skip connections to simultaneously gather data from a large number of connections. Despite the standard ReLU layer, a cropped version was investigated. The convolutional neural network (CNN) was built using a bottleneck strategy for maximum efficiency. For comprehensive data persistence, only one layer of global average pooling was used. To further enhance the detection accuracy, SVM was used in place of the softmax classifier. Compared to state-of-the-art methods, the proposed method performed better in terms of detection accuracy and computational cost in the experiments. The HILL, S-UNIWARD, and WOW context-aware steganography algorithms were tested on the BOWS2 and BOSS base datasets, validating the suggested scheme [15].