3.2.3. Multi-Resolution Segmentation (MRS)

We conducted MRS in eCognition software (version 9.4). MRS is a region-merging technique starting from each pixel forming one image object or region [22,23]. The merging criteria is local homogeneity, which describes the similarity between adjacent image objects. The merging procedure stops when all the possible merges exceed the homogeneity criteria.

MRS relies on several parameters, which are image layer weights, scale parameter (SP), shape and compactness. Image layer weights define the importance of each image layer to the segmentation process. In this research, we had three layers (RGB) in the input image. We gave them equal weights. Scale parameter is the most important parameter, which controls the average image object size [9]. A larger scale parameter allows higher spectral heterogeneity within the image objects, hence allowing more pixels within one object. Defining the proper SP is critical for MRS. In our research, we selected the SP resorting to the automatic Estimation of Scale Parameters 2 (ESP2) tool, which was advanced in [9]. Shape parameter ranges from 0 to 1. It indicates a weighting between the object's shape and its spectral color. A high value in shape parameter means less importance is put on spectral information. We set a shape parameter of 0.3 in our research. Compactness defines how compact the segmented objects are. The higher the value, the more compact the image objects may be. It was set to 0.5 in this research.
