影像驗證之研究

Abstract

隨著科技技術的創新、電腦效能的增強，在資訊化的社會中透過網際網路傳遞訊息，已成為生活中的一部份。由於透過數位化處理作影像篡改之功能日益增強，使得有心人士可輕易地經由網路截取影像並作篡改。因此如何鑑別接收到之影像是否為原始影像，遂成為一重要研究課題。由於透過網路傳輸影像多半是JPEG檔，因此本論文將提出一驗證JPEG檔案之方法。此方法之目的為鑑別影像是否遭篡改，若是則找出影像被篡改的部位，由於影像被壓縮後必產生稍微誤差，此誤差應被視為正確的部分。為了達到此目的，本論文將由透過DCT及Quantization之後的係數中找出一些特徵，這些特徵唯有在影像內容被篡改時才會改變，而在特定的重覆壓縮下仍保持不變。因此利用偵測這些特徵是否改變，即可將影像被篡改的部分找出，本論文亦將提供一些實驗結果說明此方法之有效性。

With the improvement in image-altering techniques, an image can be altered easily. Thus, a received image via the Internet may not be the original one. To treat this problem, designing a method to verify whether a received image has been altered is necessary. Since most transmitted images over the Internet are represented in JPEG format. In this thesis, a method to verify the JPEG files will be proposed. It will locate those parts with contents changed as altered ones. Those parts with a minor distortion caused by compression will be considered as correct ones. To reach this aim, a feature from the quantized coefficients of cosine transform will be introduced, and it is invariant under some restricted compression levels unless the content of the image has been altered. Based on this kind of features, the positions with contents altered can be detected. The experimental results show the effectiveness of the proposed method.