Automatic Selection of Segmentation Parameters for Object Oriented Image Classification

Image segmentation is prerequisite for object-oriented image analysis.Most image segmentation algorithms need the user to provide parameters to control the quality of the resulting segmentation.Selecting suitable parameters is a challenging task in using such algorithms.We proposed a method of parameters selection for region-growing image segmentation.Information about segmentation parameters was extracted from training sample areas of each class in the image.By multiple-segmentation of the training sample area,a maximum of objective function was found to deduce the suitable parameters for a class.Using the obtained parameters,n(the number of classes) resulting segmentations and subsequent resulting classifications were achieved.Then the n resulting classifications were fused to complete the final image classification.We tested the parameters selection for image segmentation in an object-oriented classification of remote sensing image.