ABSTRACT
Template matching is used in many applications for object identification and image segmentation because of its simplicity and stability. It frequently occurs as a normalised cross correlation between the template and patches of the image to process(Brunelli 2009). The normalised correlation coefficient m(x, y) of the template T and the patch located at coordinates (x, y) in image I is
where
I (x, y) andT (x, y) denote the brightness intensity of the pixels at coordinates (x, y) of the image and the template, respectively, while the width W and the hight H are the dimensions of the template and the imagepatch. Rotation invariance of this technique can be achieved by rotating the template to different angles, matching with all rotated templates and selecting the highest correlation coefficient.
