Improvement of watermark detection ability based on blockwise image compensation for geometrical distortions

Abstract

SUMMARYOne of the important problems of watermarking techniques is the weakness against geometrical distortions because of synchronization loss. In this paper, we propose an image compensation scheme against geometrical distortions in order to extract watermarks correctly. Our scheme collects the feature points in an image so as to recover the distortions caused by geometrical distortion attacks. Such feature points are selected by the local maximal value of partitioned regions which can be obtained by several filtering operations. Based on the computer simulated results, we have searched the optimal filtering operation and interpolation, and then studied the validity of our scheme.

Picture Watermarks Surviving General Affine Transformation and Random Distortion

“…Embedding watermarks immune to random geometric distortion requires additional methods such as searching with the help of the original picture [7], embedding watermarks with respect to the picture feature points [8,9], and embedding watermarks into constituent planes of a color picture so that they are immune to random bending [2]. To the best of our knowledge, however, no method has been reported for embedding watermarks immune to affine transformation and random distortion.…”

Use of Human Visual System to Improve Video Watermarking for Immunity to Rotation, Scale, Translation, and Random Distortion

“…Watermarking that is robust to geometric transformation has been studied in the development of various methods such as searching for watermarks embedded in pictures [13], using watermark patterns robust to rotation, scaling, and translation [10,12], and using watermark patterns suitable for estimating distortion [2]. Random distortion is one of the most difficult kinds of geometric image processing for watermarks to survive, and coping with it requires additional methods such as searching with the help of the original picture [8,9], using robust watermark patterns with specific sequences in the embedding process [15], embedding watermarks with respect to the feature points of pictures [1], and clarifying the characteristics of various types of distortion from training data [11]. The essential difficulty in treating random distortion is synchronization between the expected watermark pattern and the watermark embedded in the picture because various parts of the embedded watermark are distorted in various ways.…”