A LeNet Based Convolution Neural Network for Image Steganalysis on Multiclass Classification

Yen-Ting Chen, Tung-Shou Chen, Jeanne Chen

Abstract


It is increasingly difficult to identify data hiding techniques used in the stego-images in this era of great advancement in technology. The purpose of this paper is to identify the types of data hiding techniques of stego-images. This paper proposed the use deep learning and convolutional neural networks (CNN) to train and generate a binary classification model for prediction. This paper proposed three types of detection algorithms for stego-images. The algorithms are based on the LeNet technique of convolutional neural network (CNN) in deep learning. The first algorithm detects the DCT stego-images. The second detects the histogram stego-images. The third and final algorithm detects the LSB stego-images. Experimental results from LeNet deep learning show significantly high detection rate. The DCT stego-images showed a significant high detection accuracy of 99%, while the LSB and histograms stego-images showed 70% accuracy rate.

Keywords


Data hiding, Deep learning, Convolutional neural network (CNN), LeNet, Steganalysis


DOI
10.12783/dtcse/ccme2018/28606

Full Text:

PDF

Refbacks

  • There are currently no refbacks.