Image Forensics System Utilizing Color Filter Array and Rotation Invariant Property

• Main Authors: Cheng-Hao Liao, 廖正豪
• Other Authors: Tien-Ying Kuo
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/3x5td2

碩士 === 國立臺北科技大學 === 電機工程系研究所 === 98 === In recent years, due to the widespread use of digital imaging devices and sophisticated image editing software, it is getting easier to create and alter digital images. As a result, digital image forensics techniques become an important issue nowadays. In this thesis, we develop an image forensics system based on multiple detection techniques, including two existing literature approaches and the proposed two new methods. Our first proposed method is based on the Color Filter Array (CFA) periodic characteristics. We improve the literature Expectation Maximization (EM) algorithm to get the Maximum A Posterior Probability (MAP) for distinguishing photographic images and photorealistic computer generated images. It can also helps localize tampered image regions automatically. The second proposed method concentrates on forgery detection on duplicated region. The suspicious image is split into circle blocks, and then the concentric circle mean and Zernike moments are calculated as the feature vectors for every circle block. The feature vectors are sorted by lexicographical order, and then searched for similar pairs. Thus, the similar circle blocks can be matched by a pre-defined similarity threshold. Compared with the literature methods, our methods can improve the detection of the duplicated region forgery with any angle rotation and vertical and horizontal flipping. The experimental results and analyses demonstrate that the proposed methods are robust in forgery detection.