Abstract
Image steganography is the process of hiding secret information in digital images
while keeping the quality of the images and making it harder to find. Finding a
balance between being undetectable, having a lot of capacity, and being robust
is still a challenge in spatial-domain steganography. This thesis presents a hybrid
region of interest (ROI)-based image steganography method that integrates
texture analysis, entropy metrics, and edge detection to improve secure data
embedding in both grayscale and RGB images.
The proposed framework determines the best embedding regions by
combining entropy, texture, and edge information from Sobel and Canny filters.
This makes sure that data is only embedded in areas that are visually complex
and less likely to change. Unlike existing ROI-based or adaptive LSB techniques
that rely on a single image feature, the proposed model employs a hybrid ROI
strategy that combines entropy, texture, and edge information with correlation-guided RGB channel selection to improve imperceptibility ,payload capacity and
security.
Experimental results on the BOSSBase, UCID, and USC-SIPI datasets show
that the proposed method achieves PSNR values above 57 dB for grayscale
images and up to 66 dB for RGB images, while maintaining SSIM values close to
1.0. Compared with classical LSB and recent adaptive methods, the proposed
approach provides better imperceptibility, higher embedding capacity, and
lower detectability than traditional spatial-domain method
