Digital Image Forensics Based On Image Multi-Scale Features And Sharpening Identification

Digital imaging device has become the mainstream for the press, coming with different types of powerful and easy used image editing software on PC. This application on one hand bring great convenience to news reporters and editors to enhance the visual impact of their news images, but also led to many false news picture came out in public. Exposed in recent years, many shocked fake news events of the world happened with tampered digital image acting as their evidence, which had caused negative influence, and in another hand highlights the importance of digital image forensics.Digital image forensics technology is a new research field in image processing. This technique can determine the authenticity and integrity of digital images based on providing contents of the digital image itself, or statistical characteristics information. In this paper, based on the image contents, and by use of image invariant features under different Gaussian space scale invariant features, providing a new forensic algorithm to determine the images contains key contains copy-pasted or deleted traces. Arranged as following:firstly, a short introduction to the background will be given. Second part give the analysis and summary of common used digital image forgery tricks, highlight the steps of the splicing operation in tampering procedure. The third part describes and implements a way to extract a scale invariant feature which is called SIFT (Scale Invariant Feature Transform) algorithm. This method has strong robustness under different scale, rotation, and compression. But the traditional SIFT algorithm has large time and space consumption especially deal with high-resolution images cases, SIFT algorithm would extract excessive number descriptors which leads to high false alarm ratio. Thus in the fourth part, this paper improve its original 128-dimensional features using PCA (PCA) technology dimension reduction to 32 dimensions, which both reduces the computation time and greatly improved accuracy. Experimental results show that:the algorithm used in digital image tampering covering common means, such as blur, sharpen, after emergence, still can have a high accuracy.Focus on the tamper of USM sharpening in Photoshop, this paper presents a novel technology. After discussing the cause of edge oscillation in sharpening images, we make use of the existence of edge oscillation to extract features from step edge in images. An SVM based classifier is trained based on the extracted features for detection of sharpening images. This approach is applicable to identify the global and local sharpening images, and is robust to re-compression.