Hybrid optimization algorithm helps detect hidden messages in digital images

Research in the International Journal of Computational Science and Engineering describes a new approach to spotting messages hidden in digital images. The work contributes to the field of steganalysis, which plays a key role in cybersecurity and digital forensics.

Steganography involves embedding data within a common medium, such as words hidden among the bits and bytes of a digital image. The image looks no different when displayed on a screen, but someone who knows there is a hidden message can extract or display the message. Given the vast numbers of digital images that now exist–and that number grows at a remarkable rate every day–it is difficult to see how such hidden information might be found by a third party, such as law enforcement.

Indeed, in a sense it is security by obscurity, but it is a powerful technique nevertheless. There are legitimate uses of steganography, of course, but there are perhaps more nefarious uses, and effective detection is important for law enforcement and security.

Ankita Gupta, Rita Chhikara, and Prabha Sharma of The NorthCap University in Gurugram, India, have introduced a new approach that improves detection accuracy while addressing the computational challenges associated with processing the requisite large amounts of data.

Steganalysis involves identifying whether an image contains hidden data. Usually, the spatial rich model (SRM) is employed to detect those hidden messages. It analyzes the image to identify tiny changes in the fingerprint that would be present due to the addition of hidden data. However, SRM is complex, has a large number of features, and can overwhelm detection algorithms, leading to reduced effectiveness. This issue is often referred to as the “curse of dimensionality.”

The team has turned to a hybrid optimization algorithm called DEHHPSO, which combines three algorithms: the Harris Hawks Optimizer (HHO), Particle Swarm Optimization (PSO), and Differential Evolution (DE). Each of these algorithms was inspired by natural processes. For example, the HHO algorithm simulates the hunting behavior of Harris Hawks and balances exploration of the environment with targeting optimal solutions. The team explains that by combining HHO, PSO, and DE, they can work through complex feature sets much more quickly than is possible with a current single algorithm, however sophisticated.

The hybrid approach reduces computational demand by eliminating more than 94% of the features that would otherwise have to be processed. The stripped back information can then be processed with a support vector machine (SVM) classifier. The team says this method works better than meta-heuristic (essentially trial-and-error methods) and better even than several deep learning methods, which are usually used to solve more complex problems than steganalysis.

Research in the International Journal of Computational Science and Engineering describes a new approach to spotting messages hidden in digital images. The work contributes to the field of steganalysis, which plays a key role in cybersecurity and digital forensics.

Steganography involves embedding data within a common medium, such as words hidden among the bits and bytes of a digital image. The image looks no different when displayed on a screen, but someone who knows there is a hidden message can extract or display the message. Given the vast numbers of digital images that now exist–and that number grows at a remarkable rate every day–it is difficult to see how such hidden information might be found by a third party, such as law enforcement.

Indeed, in a sense it is security by obscurity, but it is a powerful technique nevertheless. There are legitimate uses of steganography, of course, but there are perhaps more nefarious uses, and effective detection is important for law enforcement and security.

Ankita Gupta, Rita Chhikara, and Prabha Sharma of The NorthCap University in Gurugram, India, have introduced a new approach that improves detection accuracy while addressing the computational challenges associated with processing the requisite large amounts of data.

Steganalysis involves identifying whether an image contains hidden data. Usually, the spatial rich model (SRM) is employed to detect those hidden messages. It analyzes the image to identify tiny changes in the fingerprint that would be present due to the addition of hidden data. However, SRM is complex, has a large number of features, and can overwhelm detection algorithms, leading to reduced effectiveness. This issue is often referred to as the “curse of dimensionality.”

The team has turned to a hybrid optimization algorithm called DEHHPSO, which combines three algorithms: the Harris Hawks Optimizer (HHO), Particle Swarm Optimization (PSO), and Differential Evolution (DE). Each of these algorithms was inspired by natural processes. For example, the HHO algorithm simulates the hunting behavior of Harris Hawks and balances exploration of the environment with targeting optimal solutions. The team explains that by combining HHO, PSO, and DE, they can work through complex feature sets much more quickly than is possible with a current single algorithm, however sophisticated.

The hybrid approach reduces computational demand by eliminating more than 94% of the features that would otherwise have to be processed. The stripped back information can then be processed with a support vector machine (SVM) classifier. The team says this method works better than meta-heuristic (essentially trial-and-error methods) and better even than several deep learning methods, which are usually used to solve more complex problems than steganalysis.