In this paper, we identify two current challenges\udassociated with watershed segmentation algorithms which\udattempt to fuse the visual cues of texture and intensity. The\udfirst challenge is that most existing techniques use a competing\udgradient set which does not allow boundaries to be\uddefined in terms of both visual cues. The second challenge\udis that these techniques fail to account for the spatial uncertainty\udinherent in texture gradients. We present a watershed\udsegmentation algorithm which provides a suitable solution\udto both these challenges and minimises the spatial uncertainty\udin boundary localisation. This is achieved by a novel\udfusion algorithm which uses morphological dilation to integrate\udintensity and texture gradients.Aquantitative and qualitative\udevaluation of results is provided demonstrating that our\udalgorithm outperforms three existing watershed algorithms.
展开▼
