Steganalysis Using ResNet50V2 for Deep Learning Based Hidden Data Detection

Steganography, the technology of protecting a secret message by embedding it inside a cover image, continues to be investigated and enhanced as an alternative data protection method. This paper deals with hiding multimedia files in true color RGB cover images with an emphasis on reducing the cover size, increasing hiding capacity and enhancing security of the hidden data. A proposed model (DuoHide) is presented in which a secret multimedia file, regardless of its type, is processed without un-compression, and divided between two cover images of equal size and dimensions. The multimedia file is read as a stream of bytes and split vertically into two parts, one part contains the least significant half-bytes, and the other part contains the most significant half-bytes. The two parts are hidden inside two uncompressed RGB cover images using a least significant 4-bit replacement technique. The resulting dual stego images are expected to be sent separately, through different channels, to avoid capture of both stego files by an adversary. Extraction of the secret file is achieved through merging LSB half-bytes from the two stego files. The extracted file is identical in content and structure with the original secret file. The implemented DuoHide system was evaluated using a set of public multimedia files, images, audios, and videos, of various sizes. The secret file sizes ranged from 5% to about 100% of the cover image's size. The experimental results showed that even at the highest embedding ratio, which is based on the secret-to-cover ratio, there were no perceptible visual differences between cover and stego images. The PNSR value was calculated as PSNR1, for cover1 and stego1, and PSNR2 for cover2 and stego2. The lowest PSNR value was around 31 dB for the highest embedding ratio, which is considered acceptable concerning statistical imperceptibility. The PSNR value increased as the embedding ratio decreased, reaching around 65 Decibel (dB) for the case of 5% secret-to-cover ratio. The integrity of the extracted secret file was verified through a bitwise comparison between original and extracted files, which showed zero differences. The DuoHide model is expected to provide better security for the hidden file, in case an attacker manages to capture one of the stego images and recover the hidden content because the attacker will only get an incomprehensible set of half-byte bits. An additional advantage of using a pair of stego files is that of reducing stego file size by 50%, to avoid problems and limitations of transmitting large files, especially that multimedia files are often large, and they cannot be compressed because they are already compressed. Security of the DuoHide system can further be improved by randomizing storage locations within the two stego images.

The general purpose of steganalysis algorithms is to detect a secret message in digital images. Nevertheless, previous studies have rather ignored the fact that more valuable information like message length can be extracted from stego images. In this article, a method is proposed for estimating message length in stego images resulted from PQ, which is a common steganography algorithm. In this method, the stego image resulted from PQ algorithm is first calibrated so that the corresponding cover image is obtained. Then, threshold μ is compared to the changes of global histograms of cover and stego images. If the threshold is less than the changes, first-order statistics and global histogram will be used to estimate message length; otherwise, secret message length will be estimated using second-order statistics and changes of variation of cover and stego images.

R261, Jana et al. proposed the partial reversible data hiding (PRDH) based on (7, 4) Hamming code, which deals with three images: the original image (OI), the cover image (CI), and the stego image (SI). The CI is obtained by slightly modifying OI. After that, one can embed secret into CI to generate SI. The so-called partial reversible feature is that we can reconstruct the CI faultlessly, which is almost the same to the OI. However, in Jana et al.'s PRDH, the authors only adjust redundant bits by using (7, 4) Hamming code with even (or odd) parity. This simple approach may modify 3 redundant bits at most in 7 least significant bit (LSBs) to obtain CI from OI, and this degrades the CI quality seriously. In this paper, we construct two schemes: the proposed PRDH and the modified PRDH. The proposed PRDH enhances the partial reversible property that improves the visual quality of CI. And, the modified PRDH enhances the proposed PRDH to achieve the high embedding capacity. In addition, we also point out two inaccurate descriptions in Jana et al.'s PRDH about position of embedded secret to more simplify the procedure of embedding secret.

Steganography is a process to conceal the existence of a message. Image Steganography involves hiding the message onto a cover image to obtain a resultant stego image. This paper discusses a procedure for hiding coloured message images onto coloured cover images using a lossy Image Steganography technique. The secret images undergo Discrete Cosine Transformation, quantization and AES encryption before embedding using the Least Significant Bit embedding (LSB) technique onto the cover image. Finally, Peak signal-to-noise ratio (PSNR), Structural Similarity Index measurement (SSIM), zero normalized cross-correlation (ZNCC) are used for comparing the cover and stego images as well as the extracted and original images.

This paper presents a new secret sharing scheme for identification of cheaters in the elections. In a proposed scheme secret image (voter ID) is embedded into cover image (Indian emblem) of same size and generates n stego images. The resulting stego images are visually good and look like a cover image. Various image quality metrics are used to assess the quality of stego images. Out of n stego images only last two stego images are required to reconstruct the secret and cover images without any loss. The proposed scheme offers high embedding capacity. Experimental results of proposed scheme shown to be superior in terms of PSNR and visual quality than existing schemes.

This paper presents a novel reversible data hiding method, which can recover the original image content without distortion from the stego images after the secret data have been extracted. The method utilizes the original cover image to finally generate two visually similar stego images by embedding additional data. During the data embedding process, the pixels in the first stego image are utilized to embed secret data by applying the proposed prefix matching technique, which exploits each of the gray values to carry one or two bits only with ± 1 operation. Pixels in the second stego image are adaptively modified to carry rest secret data by referring to the first stego image using the directed LSB embedding. For the recipient, the additional data can be extracted from the two stego images with the DH key, and the original cover image can be recovered with the location map. If the cover image has no boundary gray-values, one can always recover the cover image from the two stego images without the aid of any extra information. Experiments show that the proposed method can achieve desirable embedding capacity and maintain a high level of image quality.

ABSTRACTIn this paper we proposed a method for image steganography in which the cover image represented using a pseudo hexagonal structure is used for hiding the secret image. As we know, the conventional method of representing a digital image is using square-shaped pixels. Using hexagonal-shaped pixels to represent a digital image is a new approach because of its various advantages like symmetry, higher angular resolution, consistent connectivity, and higher sampling efficiency. This paper describes a method for hiding an image using a pseudo hexagonal structure cover image. In this method the conventional square pixel secret image is embedded in a cover image with pseudo hexagonal structure. Using this method we can hide an image by preserving its quality and the experimental results show that the mean square error between the secret image and the image retrieved from the stego image is zero.

Digital images are the most common cover files used for steganography. This paper investigates the pros and cons of colour and grayscale versions of images when used as steganography covers. Therefore, it comparatively evaluates the performance of both grayscale and colour versions of a given cover image when they are used with a given steganography method. Moreover, this paper examines the capability and impact of using the chrominance components for data hiding. However, two steganography methods are used as test methods, JSteg and JMQT. As a result, using colour images is better than using grayscale images for data hiding. In order to increase the hiding capacity, chrominance components can be used for data hiding. However, this degrades the quality of stego images and increases the ultimate stego image size.

A novel algorithm for colour image steganography using a new intelligent technique based on three phases

The objective of steganalysis is to detect messages hidden in a cover images, such as digital images. The ultimate goal of a steganalyst is to extract and decipher the secret message. In this paper, we present a powerful new blind steganalytic scheme that can reliably detect hidden data with a relatively small embedding rate in JPEG images as well as using a technique known as calibration. This would increase the success rate of steganalysis by detecting data in transform domain. This scheme is feature based in the sense that features that are sensitive to embedding changes are being employed as means of steganalysis. The features are extracted in DCT domain. DCT domain features have extended DCT features and Markovian features merged together in calibration technique to eliminate the drawbacks of both(inter and intra block dependency) respectively. For the lesser embedding rate, the features are considered separately to evolve a better classification rate. The blind steganalytic technique has a broad spectrum of analyzing different embedding techniques The feature set contains 274 features by merging both DCT features and Markovian features. The extracted features are being fed to a classifier which helps to distinguish between a cover and stego image. Support Vector Machine is used as classifier here.KeywordsSteganalysisDCTMarkovCalibrationSupport Vector Machine

Steganography deals with the ways of hiding communicated data in such a way that it remains confidential. Finding best position inside cover image to embed text message, maintaining a reasonable trade-off between security, robustness, higher bit embedding rate and imperceptibility are some of the challenges of steganography system. Hence, this paper presents firefly algorithm for finding best positions inside cover image in order to embed text message into cover image using Pixel Value Differencing (PVD) technique. Four different cover image was used. Experimental result showed the cover image with selected location using firefly algorithm as well as the stego image using PVD technique. The stego image was evaluated using Peak Signal to Noise Ratio (PSNR) and Mean square Error (MSE). Firefly Algorithm with PVD technique produced a promising result for image steganography.

A reversible extended secret image sharing scheme based on Chinese remainder theorem

Steganography is the art and science of hiding secret information into digital media with the intention to transmit this information. Most of the steganographic methods either use spatial domain or frequency domain for embedding the secret information. Current hybrid methods require the original cover image to extract the secret information making these methods to become not practical. This paper proposes a new blind steganographic method for digital images that combines spatial and frequency domains and does not rely on the cover image in extracting the secret information. The proposed method utilizes a chaotic map to scramble the secret information before the embedding procedure takes place. A coding map is generated during the work on spatial domain, and the original image is transformed into DWT domain, then the generated coding map is embedded in the coefficients of LL and HL sub-bands of the cover image. The drawn experimental results show that the resultant stego-images have high quality and the proposed method provides high embedding capacity compared with other methods and is robust against the visual analysis and other image processing attacks such as lossy compression and added noise. 

The multi-bit image steganography (Paul et al. Multimedia Tools and Applications 76(5):7445–7471, 2017) proposed by Paul et al. provides high embedding capacity, and adapts the embedding position of messages to image content. This paper investigates the security and robustness of multi-bit steganography. In detection resistant experiment, we embed secret messages into cover images with multi-bit image steganography to get corresponding stego images, and adopt a more popular steganalyzer (SRM) to extract features of cover images and stego images. The ensemble classifier is trained with default settings. The experimental results show detection resistance performance of multi-bit image steganography is poor. In robustness test, we embed secret messages into cover images with multi-bit image steganography to get stego images, and attack the stego images with Stirmark. Then, we extract the secret messages from the attacked stego images. The experimental results demonstrate that we can not recover the secret messages from the attacked stego images. Therefore, there are flaws in detection resistant test and robustness test of multi-bit image steganography.

The aim of this study is to hide two binary BMP images in a single BMP grayscale image. The widespread technique in image steganography is to hide one image (stego image) into another (cover image). The proposed novel method is to hide two binary images into one grayscale bitmap cover image. First of all, the proposed technique starts with transforming all grayscale cover image pixels into multiples of five using Five Modulus Method (FMM). Clearly, any modulus of five is either 0, 1, 2, 3, or 4. The transformed FMM cover image could be treated as a good host for carrying data. Obviously, it is known that the pixel value for the binary image is either 0 or 1. Therefore, by concatenating the two binary images, the composite results are 00, 01, 10 and 11. In fact, these concatenated values could be mapped using simple mapping that assigns a positive integer value such as 1 for 00, 2 for 01, 3 for 10 and 4 for 11. Consequently, a new matrix will be constructed that contains a number varying from 1 to 4 only. Fortunately, the four integer values are the same as the previously mentioned reminders of division by 5, hence, adding these four integers to the transformed FMM cover image. On the recipient side, a reverse process will be implemented to extract the two binary images. In terms of PSNR values, the cover image and the two extracted stego images have acceptable PSNR values, which yields that the proposed method is very efficient in information hiding.