Data hiding scheme with geometric distortions correction

This paper presents a new scheme for data hiding with robustness to rescale and rotation distortions. The most common geometrical distortion rescale factor and rotation factor during image manipulation can be easily estimated by comparing feature points location of original and distorted images. However, the decoder still has to have this prior information regarding the feature points of original image which is not practical. In this paper, two color spaces of RGB images can be considered as two independent channels. One is synchronization channel (SC) which transmit the prior information of original image, and another is communication channel (CC) carrying the hiding data. Contend-based image watermarking method is adopted in SC channel. The feature point extractor plays key role in our scheme. Image units are represented by Delaunay tessellations constructed from extracted feature points. The scale factor and rotation angle estimated from feature points are parsed into binary data as synchronization information and embedded to each image unit. The synchronization information can be extracted successfully if at least two image units are robustness against the distortion. Consequently, the rescaling factor and the rotation angle can be estimated and corrected. Several key problems such as the feature point detector improvement, image unit design and SC synchronization information hiding procedure are discussed in details. The experiment results and discussions are given as well.