Image thresholding based on template matching with arctangent Hausdorff distance measure

This paper proposes a relatively simple yet highly effective thresholding method based on the template matching idea. A reference template is first produced by sampling edge pixels of an input gray level image. A series of input patterns are sequentially obtained by extracting the boundaries of objects in the binary version of gray level image, where the binary images are sequentially generated by thresholding the input gray level image with each possible gray level from 0 to 255. A newly proposed arctangent Hausdorff distance (AHD) measure is applied to estimate an input pattern E-opt that is most similar to the reference template. The binary image corresponding to the input pattern E-opt is then regarded as the final thresholding result. Experimental results show that the segmentation quality of the proposed approach surpasses 11 existing thresholding methods. (C) 2013 Elsevier Ltd. All rights reserved.