Influence of shape parameters on optimal scale selection in multi-resolution segmentation

Multi-segmentation includes three categories: weights of bands, segmentation scale (f) and shape parameters, in which f is the most important parameter. Lots of methods have been done on how to find the optimal segmentation and have been applied to different situations. But most of them didn´t consider the impact of shape parameters on the optimal scale. This paper experimentally proves that the shape parameters would impact the optimal scale selection. Based on the sampled values of the above three parameters we got 26 settings, each of which have 10 segmentation scales. Through experimental analysis, the following preliminary conclusion can be drawn: (1) the mean standard deviation (MSD) of all the bands and their mean brightness pointed to the same optimal scale; (2) the two shape parameters have different impacts on the selection of optimal scale; (3) a greater weight of shape will help improve the efficiency of segmentation conditionally to the selection of optimal scale.