Tampered Image Detection Using Image Matching

With the development of image processing software, such as Photoshop, scientific analysis for detecting tampered images becomes a critical research issue. A hybrid image matching algorithm for image analysis is proposed and evaluated. Features are extracted using blob detectors and interest point detectors. This combination ensures sufficient correspondences in small image patches while maintaining high accuracy. Nearest neighbors are searched using hybrid spill tree in order to reduce the computational load. Applying the algorithm to tampered image detection, a novel solution is proposed utilizing the scale space information and gradient orientation information from the corresponding features. Furthermore, an order regularity of scale space between the corresponding patches and a description for object rotation are also proposed. In addition, we conducted experiments on tampered images, proving the method to be powerful.