A novel template matching algorithm based on the contextual semantic information

This paper presents a novel template matching algorithm that copes with the occlusion on the target image. Most of the previous methods are sensitive to the occlusion because it induces some patches with similar semantic concepts but distinct appearances. Thus the template matching algorithm hardly accomplishes the task based on the appearance similarity only. To overcome this limitation, we integrate the contextual semantic information into the template matching algorithm. To this end, we first segment the template image into 9 patches. The center patch is used to compute the appearance similarity and its neighborhood patches are adopted to construct the contextual semantic constraint. And then we obtain the integrated distance by introducing the pseudo-likelihood to combine the feature appearance similarity and contextual semantic information together. Finally, the arbitrary regions of a target image are matched with the template image via integrated distance. The experimental results demonstrate the proposed method is more robust to occlusion than previous template matching techniques.