Robust Reversible Watermarking via Invariant Image Classification and DHS

In this paper a new reversible watermarking scheme is proposed based on Dynamic Histogram Shifting (DHS). It adaptively takes care of the local speci?cities of the image content. Further a reference image is derived which is a copy of image itself. This reference image is invariant to the watermark insertion process. For identifying parts of the image that can be watermarked, Classi?cation process is used. Further, Embedder and Extractor both use same reference image. Hence synchronisation exists between embedder and extractor for message extraction and image reconstruction. For sensitive images like medical or military images, our scheme can insert more data with lower distortion than other methods.