A New Image Segmentation Method Using Clustering and Region Merging Techniques

Abstract

Clustering technique is one of the commonly used image segmentation methods. However, all of the clustering existing techniques require user-specific parameters as input. The result of segmentation depends on these parameters; therefore, the selection of an optimal value of these parameters is very crucial. In this paper, we have introduced image segmentation method based on new clustering algorithm where there is no need for initialization of the user-specific parameter. The proposed clustering technique is based on the density estimation of the surrounding pixel values. The recursive approach is used for the density estimation. After the segmentation using clustering, small segments can be present. The presence of the small segments may result in lower performance of the segmentation output. Therefore, each of the segments is merged with another neighbor segment by choosing the best matching segment. For every iteration, one of the segments will be merged with another segment and the iteration with the best performance will be the final segmentation output. The proposed method is compared with the other clustering-based segmentation methods by calculating the performance evaluation indices, and it validates the effectiveness of the proposed algorithm.