A robust image identification using trace-Hausdorff combination

Large numbers of image databases now exist that contain multiple modified versions of the same image. An extreme example of this is the large number of modified versions of images on the Internet (Web site). There is a need to develop tools that will enable the identification of all of the original and modified versions of the same images. Thus, in this paper proposes a robust method for image identification with variant illumination, compression, flip, scaling, rotation and gray scale conversion. Techniques introduced in this work are composed of two stages. First, the signature of image is to be detected by the trace transform. Then, in the second stage, the Hausdorff distance and modified shape context are employed to measure and determine of similarity between models and test images. Finally, our method is evaluated with experiments on a set of over 60,000 unique images and one billion image pairs. The extensive experimental results show detection rate of over 83% at false-positive rate below 1 per million.