Detecting Partial Duplications in an Image

—The Improvement of Estimating Rotation Angle and Reversal—

Abstract

We present a method for improving the accuracy of detecting the duplicated regions, accompanied with rotation and/or reversal, in an image. In judgment of reversal and estimation of rotation angle, we add the following processes, 1) paring pixels by principle component values, 2) calculating gradation centroid vectors for each of paired pixels, 3) evaluating the difference in angles and magnitudes between the centroid vectors. We conducted an experiment by using 100 duplicated images. The proposed method gave the correct estimation for 90 images, while the conventional one for 68 images.