This paper introduces a class of rank-order-based template matching criteria that are multiplier-free and independent of the dc variations of the image. The core component of these criteria is a grayscale morphological hit-or-miss transform (GHMT). Experimental results show that the GHMT features sharp and robust indications in the presence of Gaussian noise. The idea of the GHMT is used to develop more general forms of matching criteria that are robust to both Gaussian and impulsive noises.

The paper extends Kosravi and Schafer (1994) which introduced the notion of normalized min correlation (NMC) as a new criterion for template matching applications. The NMC features low complexity and robustness with respect to Gaussian noise. It is also independent of the local average of the image, which is a desired property in many template matching applications. However, its sensitivity to impulsive noise increases with the size of the template. The paper introduces two different modifications of the NMC which are robust with respect to both Gaussian and impulsive noise. The performance of the new modifications are examined in motion estimation applications. As a different way of looking at the NMC, the notion of grayscale hit-or-miss transform is introduced. This is an extension of the binary hit-or-miss transform to grayscale.>