<title>Morphological approach for thresholding noisy images</title>

Abstract

Image segmentation is an important pre-processing step before object recognition. Here, we assume that an image consists of three main primitives, namely the noise, the object and the varying background. First we shall show a mean to characterize the sizes of these primitives based on the morphological opening. Second, we investigate how an image can be effectively enhanced by looking for blocks inscribed under the image surface and then removing the top of the noisy background and the bottom of the foreground, which is small speck noise, constructed from the surfaces of the inscribed blocks. With these findings, a morphological segmentation algorithm is thus formulated. Experimental results are included to illustrate its superiority over other two segmentation algorithms.