Ordering pixels for fast ZNCC template matching

Basic-mode partial correlation elimination (BMP) has received popularity in recent years for fast template matching based on Zero-mean Normalized Cross Correlation (ZNCC). BMP is a full search equivalent template matching method, which examines pixel by pixel, and eliminates the rest of the computation as soon as it finds out that the current location is not likely to be the best match. In this paper, we reveal that some pixels have more contributions for matching than others. Based on this observation, an optimal ordering method is proposed to examine the pixels to accelerate the computation of BMP, called Ordering BMP (OBMP) for short. To further accelerate OBMP, we present a near-optimal ordering method which is solely based on the template image, thus can be generated offline. Extensive experiments show that the optimal or the near-optimal OBMP is on average 2.36 times faster than BMP for a template with 50×50 pixels.