This paper presents an improved segmentation approach derived from mean shift for
natural images. The optimal color bandwidth under Plug-in rule is not always satisfying
in the actual vision tasks, and a changing color bandwidth is helpful for controlling
the segmentation result. The performance of direct density searching is better
than mean shift under the same spatial bandwidth. A global optimization criterion for
mode merging stabilizes the result in segmenting different images. Merging of texture
regions mostly eliminates the influence from texture features. Based on the adjustable
color bandwidth, direct density searching and a global criterion, the improved clustering
approach performs better than mean shift, shown from experimental results. In
addition, an image is partitioned into some local patches after mode detection. These
patches can be taken as the initial segmentation for further processing that is based on
a global optimization criterion with texture or statistical features.