在許多的浮水印嵌入系統中,浮水印的取回原理主要建立在已知浮水印嵌入時的嵌入位置。然而幾何失真會重新排列浮水印的嵌入位置,因而導致浮水印取回的失敗。本篇論文所提出的機制便是針對解決幾何失真中的旋轉、縮放、平移為目標。 所提出的機制包含了兩階段:私有分享生成及秘密取回。在私有分享生成階段,該機制會經由RST-不變量的領域取得影像特徵,再根據秘密分享機制去產生相對應的私有分享以供驗證。在秘密取回階段,同樣於RST-不變量的領域取得影像特徵,與所保存的私有分享結合即可取回秘密影像來做著作權的驗證。 從實驗中顯示這個機制可以抵抗下列幾種影像處理:平均模糊、對比調整、高斯模糊、JPEG (失真壓縮)、中值濾波、亮度調整、銳利化、均勻雜訊掺入、變動雜訊掺入、輕微的仿射轉換、輕微的擠壓、輕微的擴張、縮放、像素行列移除、旋轉、旋轉混和輕微裁剪、旋轉混和縮放、等化、列印再掃描。
In many watermarking systems, the main principle to retrieve the watermark is based on the knowledge of the positions for embedding the watermark. However, the geometric distortions rearrange the watermark embedding positions, and this often results in the failure of the watermark retrieval. In this thesis, the proposed scheme aims at resisting the rotation, scaling, and translation of the geometric distortions. The scheme contains a private share generation phase and a secret image retrieval phase. In the private share generation phase, the scheme employs the RST-invariant domain to obtain the feature share of the host image, and then generate the corresponding private share based on the secret sharing scheme for authentication. In the secret image retrieval phase, the feature share of the suspect image is retrieved from the RST-invariant domain, and then combined with the private share to regain the secret image for copyright authentication. The experimental results show that the scheme can withstand several attacks such as averaging, contrast adjustment, Gaussian blurring, JPEG compression, median filtering, Brightness-adjustment, sharpening, uniform noise adding, variance noise adding, slight affine transformation, slight pinching, slight punching, rescaling, lines removing, rotating, rotating with cropping, rotating with scaling, equalization, and print-and-scan.