Decaying extension based phase correlation for robust object localization in full search space

Phase correlation is an efficient tool for precisely localizing an object of interest in an image. However, its performance is severely deteriorated, if: (1) the images contain significant mismatch between the intensity levels of the pixels at the opposite boundaries, (2) the object in the search image is slightly rotated or scaled, or (3) the search image is significantly noisy or blurred. Some of these problems have been addressed by previous techniques, but at the cost of an spatial constraint that the object is fairly inside some central region in the search space. Therefore, we propose an efficient and effective preprocessing technique, that extends the search image and the template with new pixels having smoothly decaying values. It is demonstrated that the proposed method outperforms two recent techniques in localizing an object of interest in real images, especially when the object lies away from the central region in the search space.