Scholarlink

<p>Steganography has become a popular method for secret communication in MP3, a compressed music format. MP3 audio is widely used by users, and its transparency in hiding information makes it useful for secret communication. We present a novel steganalysis framework, MFCCResNet, that is especially locating concealed data in compressed audio in order to satisfy the increasing demand for reliable forensic detection. Existing investigations depend upon characteristics like spectrograms and Modified Discrete Cosine Transform (MDCT) coefficients, although both methods have their own disadvantages. MDCT features provided less effective in detecting the small distortions introduced by steganography techniques and Spectrogram based features affected from sensitivity to small change and high dimensionality while choosing the parameters, which can cause overfitting, especially when data are limited. So, the feature in the proposed system uses Mel-Frequency Cepstral Coefficients (MFCCs) provide a compact and perceptually grounded representation aligned with human auditory processing. Deep learning models integrated with MFCC feature has challenges such as vanishing gradients when scaled to greater depth. To overcome this, our framework incorporates a Deep Squeeze and Excitation Residual Network (SE-ResNet), whose shortcut connections allow stable gradient flow and support the extraction, discriminative patterns across many layers. By combining MFCC features with Squeeze and Excitation ResNet’s residual learning, the proposed MFCCResNet model delivers a rich and scalable solution for MP3 audio steganalysis, achieving substantial improvements in accuracy, precision, recall, and F1-score compared with existing methods.</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

MFCCResNet: A Deep Squeeze and Excitation Residual Network with MFCC Features for Mp3 Steganography Detection

  • 期刊论文

  • 发表日期：

    2026年4月10日

查看全文
  • J. Stella,
  • P. Karthikeyan

摘要

Steganography has become a popular method for secret communication in MP3, a compressed music format. MP3 audio is widely used by users, and its transparency in hiding information makes it useful for secret communication. We present a novel steganalysis framework, MFCCResNet, that is especially locating concealed data in compressed audio in order to satisfy the increasing demand for reliable forensic detection. Existing investigations depend upon characteristics like spectrograms and Modified Discrete Cosine Transform (MDCT) coefficients, although both methods have their own disadvantages. MDCT features provided less effective in detecting the small distortions introduced by steganography techniques and Spectrogram based features affected from sensitivity to small change and high dimensionality while choosing the parameters, which can cause overfitting, especially when data are limited. So, the feature in the proposed system uses Mel-Frequency Cepstral Coefficients (MFCCs) provide a compact and perceptually grounded representation aligned with human auditory processing. Deep learning models integrated with MFCC feature has challenges such as vanishing gradients when scaled to greater depth. To overcome this, our framework incorporates a Deep Squeeze and Excitation Residual Network (SE-ResNet), whose shortcut connections allow stable gradient flow and support the extraction, discriminative patterns across many layers. By combining MFCC features with Squeeze and Excitation ResNet’s residual learning, the proposed MFCCResNet model delivers a rich and scalable solution for MP3 audio steganalysis, achieving substantial improvements in accuracy, precision, recall, and F1-score compared with existing methods.