Scholarlink

<p>Artificial Intelligence (AI), specifically deep learning methods such as Convolutional Neural Networks (CNNs), greatly advances digital image steganography by allowing more complex and secure embedding of hidden information within images. This method supports secret communication by inserting confidential data—such as messages, passwords, or sensitive content into normal-looking images without drawing attention. Unlike cryptographic methods, steganography hides information without encrypting it or relying on keys. Leveraging AI, particularly CNNs, improves the ability to embed hidden data with greater capacity, minimal visual distortion, and stronger resistance to detection by steganalysis techniques. Nonetheless, significant challenges persist in balancing between large data embedding and preserving image integrity, addressing differences in cover image sources, avoiding quality loss in stego images, and ensuring adaptability across various image types, all essential to effectively conceal the secret information while avoiding detection. Thus, a novel AI-based model is designed for digital image steganography. Initially, the required images were collected from standard datasets. Then, the collected images are given into the steganography detection phase, where it employs Dual attention-based Adaptive Dense-Atrous spatial pyramid pooling Encoder-Decoder Network (DADA-EDNet). Here, Atrous Spatial Pyramid Pooling (ASPP) are applied for efficiently extracting the features from the images, enabling robust and accurate embedding of covert information while reducing the deformation, and enhancing the hiding ability, rendering it more complex for others to recognize the embedded covert details. Incorporating encoder-decoder models aids in preserving the visual reliability of cover images. Moreover, to enhance detection efficacy, several parameters of the DADA-EDNet are tuned using the Random Attribute Updated Human Evolutionary Optimization (RAU-HEO). Finally, the developed DA-ADASPP-EDNet offers the detected outcome by using the RAU-HEO. Then, experimental validation is performed on the developed model to prove its effectiveness over existing models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Artificial intelligence-based digital image steganography model using dual attention mechanism with adaptive dense-aspp encoder-decoder architecture for improved security

  • 期刊论文

  • 发表日期：

    2026年4月2日

查看全文
  • Ajay Kumar Patwari,
  • G. Venkata Rami Reddy

摘要

Artificial Intelligence (AI), specifically deep learning methods such as Convolutional Neural Networks (CNNs), greatly advances digital image steganography by allowing more complex and secure embedding of hidden information within images. This method supports secret communication by inserting confidential data—such as messages, passwords, or sensitive content into normal-looking images without drawing attention. Unlike cryptographic methods, steganography hides information without encrypting it or relying on keys. Leveraging AI, particularly CNNs, improves the ability to embed hidden data with greater capacity, minimal visual distortion, and stronger resistance to detection by steganalysis techniques. Nonetheless, significant challenges persist in balancing between large data embedding and preserving image integrity, addressing differences in cover image sources, avoiding quality loss in stego images, and ensuring adaptability across various image types, all essential to effectively conceal the secret information while avoiding detection. Thus, a novel AI-based model is designed for digital image steganography. Initially, the required images were collected from standard datasets. Then, the collected images are given into the steganography detection phase, where it employs Dual attention-based Adaptive Dense-Atrous spatial pyramid pooling Encoder-Decoder Network (DADA-EDNet). Here, Atrous Spatial Pyramid Pooling (ASPP) are applied for efficiently extracting the features from the images, enabling robust and accurate embedding of covert information while reducing the deformation, and enhancing the hiding ability, rendering it more complex for others to recognize the embedded covert details. Incorporating encoder-decoder models aids in preserving the visual reliability of cover images. Moreover, to enhance detection efficacy, several parameters of the DADA-EDNet are tuned using the Random Attribute Updated Human Evolutionary Optimization (RAU-HEO). Finally, the developed DA-ADASPP-EDNet offers the detected outcome by using the RAU-HEO. Then, experimental validation is performed on the developed model to prove its effectiveness over existing models.