Using noise models to estimate rank parameters for rank order greyscale hit-or-miss transforms

The Hit-or-Miss Transform (HMT) is a morphological transform which can be used for template matching and other applications. Recent developments of the HMT include extensions of the transform for application to greyscale images as well as a variety of techniques aiming to improve its noise robustness. One popular technique for improving noise robustness is to use rank order filters in place of the traditional morphological operations of erosion and dilation. However, very few authors give consideration to developing generic techniques for estimating the rank parameters they introduce. Very recently, techniques which use ROC curves, or the SEs designed for object detection, have been presented for estimating optimal values for the rank parameter. This paper presents a new, simpler technique which uses noise models extracted from the image set under study to estimate the optimal rank parameter.