New image forensic tool detects forgery and manipulation

A new image forensic tool that investigates color sensor pattern noise in images and uses Zernike moments of small image demonstrates cutting-edge technology in both color modification and copy-move detection.

Article  |  Fall 2015

Although color modification is a common forgery technique, there has been no reported forensic method for detecting this type of manipulation thus far. However, a research team led by Prof. Heung-Kyu Lee has developed a novel algorithm for estimating color modification in images acquired from digital cameras when the images have been modified. Most commercial digital cameras are equipped with a color filter array (CFA) for acquiring the color information of each pixel. As a result, the images acquired from such digital cameras include a trace from the CFA pattern. This pattern is composed of the basic red, green, blue (RGB) colors, and it changes when color modification is carried out on the image.

An advanced intermediate value counting method was designed for measuring the change in the CFA pattern and estimating the extent of color modification. The presented method is verified experimentally by using 10,366 test images. The results confirm the ability of the proposed method to estimate color modification with high accuracy.

Another forensic technique was also developed to localize duplicated image regions based on the Zernike moments of small image blocks. Rotation invariance properties are exploited to reliably unveil duplicated regions after arbitrary rotations. A novel block matching procedure was devised based on locality sensitive hashing and reduce false positives by examining the moments’ phase. A massive experimental test was conducted against state-of-the-art methods under various perspectives, examining both pixel-level localization and image-level detection performance. By taking signal characteristics into account and distinguishing between “textured” and “smooth” duplicated regions, the presented method outperforms existing methods, particularly when the duplicated regions are smooth. Experiments indicate high robustness against JPEG compression, blurring, additive white Gaussian noise, and moderate scaling.

Based those forensic technologies, a service platform was developed for digital image forgery detection, and the service has been provided online to the public since June 8, 2015 at http://forensic.kaist.ac.kr. The service system can also determine other various forgeries, including blurring, re-sampling, copy rotation move, global double JPEG, local double JPEG, splicing, and some hybrid modifications. Anyone wishing to test an image can upload it to the website and obtain results within a minute. This kind of image forgery detection service is the first of its kind in Korea.

Prof. Heung-Kyu Lee explained, “The research team members have been quite surprised to find that several thousands of images are being uploaded every day, and the web server frequently stops due traffic overloads. Some institutes, including police stations, law firms, newspapers, and medical institutions, are asking to investigate whether their images are forged or not. Our researchers realized that more developed technologies are required to solve many issues from those requests. A 2.0 version will be possible in near future with more sophisticated technologies.”

Over the past few years, the research team has published numerous papers in journals in the image forgery area, including Forensics Science International, IEEE Trans. on Information Forensics and Security, Sensors, Pattern Recognition Letters, and Electronics Letters.