Color Image steganography using Deep convolutional Autoencoders based on ResNet architecture
Abstract: In this paper, a deep learning color image steganography scheme combining convolutional autoencoders and ResNet architecture is proposed. Traditional steganography methods suffer from some critical defects such as low capacity, security, and robustness. In recent decades, image hiding and image extraction were realized by autoencoder convolutional neural networks to solve the aforementioned challenges. The contribution of this paper is introducing a new scheme for color image steganography inspired by ResNet architecture. The reverse ResNet architecture is utilized to extract the secret image from the stego image. In the proposed method, all images are passed through the prepossess model which is a convolutional deep neural network with the aim of feature extraction. Then, the operational model generates stego and extracted images. In fact, the operational model is an autoencoder based on ResNet structure that produces an image from feature maps. The advantage of proposed structure is identity of models in embedding and extraction phases. The performance of the proposed method is studied using COCO and CelebA datasets. For quantitative comparisons with previous related works, peak signal-to-noise ratio (PSNR), the structural similarity index (SSIM) and hiding capacity are evaluated. The experimental results verify that the proposed scheme performs better than traditional and pervious deep steganography methods. The PSNR and SSIM are more than 40 dB and 0.98, respectively that implies high imperceptibility of the proposed method. Also, this method can hide a color image of the same size in another color image, which can be inferred that the relative capacity of the proposed method is 8 bits per pixel.
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.
Top Community Prompts
Explain it Like I'm 14
Overview
This paper is about hiding one color image inside another color image so that no one can tell it’s there. This is called steganography. The authors use deep learning (a kind of artificial intelligence) to do this in a way that keeps the hidden image safe, the visible image looking normal, and makes it easy to get the hidden image back later.
What questions does the paper try to answer?
- How can we hide a full color image inside another color image without anyone noticing?
- Can we make the hidden image easy to recover with high quality?
- Can we use a single smart design so that the “hiding” part and the “revealing” part work the same way, making the system simpler to build and train?
How did they do it?
Think of it like digital “invisible ink” for pictures:
- They built a two-part AI system:
- A preprocess model: This is like a scanner that looks at an image and pulls out important features (edges, textures, patterns) instead of using raw pixels. It turns each image into a compact “feature map” that’s easier to work with.
- An operational model: This is like a painter that takes feature maps and produces whole images. It’s an autoencoder based on ResNet.
What those terms mean in everyday language:
- Autoencoder: Imagine a smart copier that learns to compress an image into a small summary and then recreate the image from that summary.
- ResNet (Residual Network): A design trick that adds “shortcuts” through the network so information doesn’t get lost as it goes deeper. Think of it as leaving sticky notes along a long pathway so you don’t forget important details.
How the hiding and revealing works:
- Hiding (embedding phase): The secret image and the cover image are both passed through the preprocess model to get their feature maps. These maps are merged into one big map. The operational model then “paints” a normal-looking stego image (the cover image with the secret hidden inside).
- Revealing (extraction phase): The receiver takes the stego image, passes it through the same preprocess and operational models, and reconstructs the secret image.
A key idea: The same operational model is used for both hiding and revealing. Because the preprocess step turns inputs into the same-sized feature maps, the hiding and revealing models can be identical. This makes designing and training the system simpler and more reliable.
Training and datasets:
- They trained on two common image collections: COCO (many everyday scenes and objects) and CelebA (many face images).
- The system is trained end-to-end, meaning it learns to balance two goals at once: keeping the stego image looking like the cover image and reconstructing the secret image clearly.
- They used a standard “difference” measure called MSE (Mean Squared Error), which is basically “how far off are we on average per pixel,” and a weight to trade off between image quality of the stego image and the extracted secret.
What did they find?
The results were strong:
- Image quality: Using measures called PSNR and SSIM (ways to judge how similar two images look), the stego images looked very close to the cover images, and the extracted images looked very close to the original secret images.
- PSNR (Peak Signal-to-Noise Ratio): Higher is better. Their results were often around or above 40 dB, which means the differences are tiny and hard to see.
- SSIM (Structural Similarity): Closer to 1 is better. Their results were typically above 0.98, meaning the images look almost identical in structure.
- Capacity: They can hide a full color secret image of the same size as the cover image. In simple terms, that’s like fitting one complete 256×256 color image inside another 256×256 color image without making it look suspicious. They report a relative capacity of 8 bits per pixel, which is very high.
- Security: Tests that try to detect hidden information (like looking at image histograms or using steganalysis tools) found it hard to spot the differences. This suggests the method is stealthy and secure.
Why does this matter?
- Better secret sharing: This kind of method can help people or organizations share sensitive images safely without drawing attention, useful in secure communication.
- High quality and easy recovery: The hidden image can be brought back with high quality, which means important details aren’t lost.
- Simpler, more reliable design: Using the same model for hiding and revealing makes the system easier to build, tune, and trust.
Summary and impact
This paper shows a practical and effective way to hide full-color images inside other images using deep learning. By combining a feature-extracting preprocess model with a ResNet-based autoencoder, they achieve high visual quality, strong security, and a very high hiding capacity. This advances the field of image steganography and could be used for safer digital communication. As with any powerful tool, it should be used responsibly, since hidden data can be used for good (privacy) or misuse if handled improperly.
Collections
Sign up for free to add this paper to one or more collections.