A multi-feature based invertible authentication watermarking for JPEG images

Invertible authentication techniques have been used in some special applications (such as medical or military applications), where original data can be recovered after being authenticated. In this paper, we proposed a new scheme combine features (relation of coefficients) in DCT domain with content features (edges or textures) of JPEG image into multi-feature, which is useful to authenticate images in both spatial and frequency domains. A novel invertible data embedding mechanism based on random walk is proposed. Experimental results show that the algorithm is sensitive to intended tampers while accepting appropriate JPEG recompressions, and even more, it has the ability to locate corrupt areas. Compared with authentication scheme based on features extracted from single domain, the "miss alarm" probability of our scheme is much lower.