Toward stronger energy compaction for high capacity dct-based steganography: a region-growing approach

Abstract

It has been proven that the strong energy compaction property of the Discrete Cosine Transform (DCT) has a relation with the inter-pixel correlation level in the spatial domain of the cover image. Recent studies have proven that the higher the correlation between pixels in the spatial domain, the stronger the energy compaction property of the DCT. Therefore, several state-of-the-art image hiding schemes aim to increase the homogeneity by segmenting the cover image into homogeneous segments to exploit the strong compaction property of the DCT. Early attempts had adopted the idea of segmenting the cover image with fixed-sized blocks to increase the homogeneity. In other attempts, the homogeneity was increased by utilizing the quad-tree segmentation technique, which showed improved results in hiding capacity and stego quality compared to the fixed-block based hiding scheme. This paper proposes the Adaptive Region-Growing (ARG) image hiding scheme, which aims to maximize the inter-pixel correlation level using the region-growing segmentation method. This segmentation technique has the ability to precisely segment the cover image into free-shaped homogeneous regions. Since objects in an image are usually free-shaped regions, it is expected that segmenting into free-shaped regions will maximize the homogeneity level within a region more than block-based segmentation techniques. Therefore, experimental results have shown superior performance of the proposed ARG scheme over competitive steganography techniques in both the hiding capacity and the quality of the stego image that reached up to 40.36 dB at 22.34 bpp.