Abstract:
Objectives: This paper intends to propose a secure steganography approach in JPEG compressed domain by providing more possibilities to analyzing the DCT coefficients in lower frequency area by modifying the primary Quantization Table (QT) with generating random data hiding patterns. Methods/Statistical analysis: The upper left part of the primary QT extracted from gray scale image dataset is modified by multiplying the factors 1⁄4, 1⁄2 and 3⁄4 to produce secondary QTs for investigating randomly generated data hiding patterns in lower frequency area of quantized DCT coefficients by Least Significant Bit (LSB) method. We create a pool of QTs and those tables are cross checked with randomly generated hiding patterns to find best QT with appropriate data hiding pattern by assessing Peak Signal to Noise Ratio (PSNR). Our method can be used to attain trade-off between some parameters such as image features, QT, data hiding pattern. Further, statistical features of a given image data set are extracted and analyzed with the selected QT and appropriate hiding pattern by using R software. Findings: The Experimental results revealed that our proposed method can embed high capacity data (57 bits per block) without noticeable visual artifacts by considering lower frequency coefficients for data hiding by assessing the image steganographic requirements. The maximum PSNR value 48 and the minimum PSNR value 32 are found among the fifty jpeg gray images based on their contents. Although, the embedding capacity and PSNR fluctuate among images, our method can be used to attain trade-off between some parameters such as image features, QT, data hiding pattern. Further, statistical features of a given image data set are extracted and analyzed with the selected QT and appropriate hiding pattern using R. The hypothesis test was deployed among the QTs, hiding patterns and image features. The following five P-Values, 0.04128, 0.02486, 0.02241, 0.04898, 0.01966 less than 0.05 show the good relationship between the QTs, hiding patterns and image features among the cover images and also the following four P-Values, 0.00685, 0.03017, 0.001085, 4.568e-12, less than 0.05, show the good relationship between those above mentioned factors among the stego images. Application/Improvements: Finally, we present a secure model to explore the relationship between QT, hiding pattern and image contents. The found model is stego invariant for sender and receiver that will enable them to identify
QT and pattern by extracting the image features without fully decoding the stego jpeg image. This method is very practical and adaptable for extending QTs and hiding patterns.
