Research Article
BMP Color Images Steganographer Detection Based on Deep Learning
@INPROCEEDINGS{10.1007/978-3-030-89814-4_44, author={Shuaipeng Yang and Yang Yu and Xiaoming Liu and Hong Zhang and Haoyu Wang}, title={BMP Color Images Steganographer Detection Based on Deep Learning}, proceedings={Mobile Multimedia Communications. 14th EAI International Conference, Mobimedia 2021, Virtual Event, July 23-25, 2021, Proceedings}, proceedings_a={MOBIMEDIA}, year={2021}, month={11}, keywords={WiserNet BMP color image Steganographer detection DBSCAN}, doi={10.1007/978-3-030-89814-4_44} }- Shuaipeng Yang
Yang Yu
Xiaoming Liu
Hong Zhang
Haoyu Wang
Year: 2021
BMP Color Images Steganographer Detection Based on Deep Learning
MOBIMEDIA
Springer
DOI: 10.1007/978-3-030-89814-4_44
Abstract
A user who achieves covert communication by embedding secret information in the original image is called steganographer. Steganographer detection determines which user sent a secured image with a secret message. Existing steganographer detection algorithms take gray images as the main research content. To better adapt to the reality, we propose a WiserNet-based steganograph detection algorithm for the characteristics of BMP color images, and the process is divided into the following three steps: feature extraction through each channel convolution structure, prevent the conventional convolution structure destroy the correlation between the color image channel operation, reduce the number of the extraction of feature dimension. The use of a per-channel convolution structure makes it easier to extract color image features, and the low-dimensional feature vector reduces the time required for subsequent clustering algorithms, which improves the efficiency of steganographer detection. Simulation experiments are conducted for the classification of feature extractors, detection of different steganographic rates, and detection of different image scales. First, the steganalysis binary classification results of this algorithm are compared with similar algorithms, and the classification accuracy is 84.90% when the steganalysis rate is 0.4 BPC, which is 1.11% higher than Ye-Net and 0.83% higher than Xu-ResNet. Since there is very little published research on steganography detection of color images, four feature extractors, Ye-Net, Xu-ResNet, SRNet, and WiserNet, will be used in this experiment to replace the WiserNet-100 feature extractor in the steganography detection algorithm. The results show that the detection accuracy of the algorithm proposed in this paper reaches 93% when the embedding rate is 0.2 BPC, and the detection accuracy reaches 100% when the embedding rate is greater than 0.2 BPC. The steganographic detection accuracy reaches 84% when the graph scale is 60% and the steganographic rate is 0.2 BPC. In terms of detection time, the WNCISD-100 is 7.79 s, which is 50% less time-consuming compared to SRSD.
