Scale selection for morphological top-hat transformation based on mutual information

Multi-scale morphological top-hat transformation has been widespread used in pattern recognition, but the selection of the scale is still very subjective and empirical. To automate the scale selection, an automatic scale selection approach based on mutual information is proposed and the property of the proposed criterion based on mutual information is investigated extensively in this paper. Experimental results indicated that the proposed scale selection algorithm was capable of selecting the proper scale to the top-hat transformation for the extraction of uneven background illumination.