Mean-Shift Algorithm for Segmentation
Erik Cuevas, Alma Rodríguez · 2024
The use of several pixel characteristics has confirmed its better performance in comparison with approaches based on only intensity information. The mean-shift (MS) scheme corresponds to a clustering method that has been extensively used for segmentation purposes. Although MS maintains interesting segmentation results, its operation presents a high computational overload. This fact makes its use complicated in schemes where the number of pixel characteristics is more than two. In the present chapter, we discuss the way in which the MS algorithm is employed for segmentation purposes. In the chapter, it is considered that these are three descriptive characteristics that involve the information of the intensity value, the non-local mean, and the local variance of each pixel in the image. To reduce the computational procedure, the MS method is modified to operate by using only a representative group of pixel characteristics. For this purpose, two sets of data are produced: operative elements (the reduced data employed in the MS process) and inactive elements (the remainder of the accessible data). In contrast to the original MS scheme, which considers Gaussian models, in this chapter, the Epanechnikov kernel function is used for determining the density points of the feature map. The results obtained by the MS method with the operative data are then used to involve the inactive data. Under this process, each inactive pixel is designed to the cluster corresponding to the nearest operative pixel. As a final step, clusters maintaining the minimal number of elements are blended with other nearby clusters.