Multi-scale segmentation algorithm is the basis for classification and information extraction of object-oriented image analysis. Due to no obvious mathematical relationship between the scale parameter and the success of the segmentation, therefore, the selection of parameters highly depends on the user's experience. Users select parameters by trial and error method, which is iterative and time-consuming. The international famous object-oriented image analysis software eCognition also has not solved this problem. Aiming at the selection of multi-scale segmentation algorithm parameters (scale, shape, etc), this paper makes use of self-organizing, adaptive and self-learning characteristics of evolutionary computation to automatically optimize the parameters of the multi-scale segmentation algorithm according to the evaluation of segmentation results. This method eliminates blindness and subjectivity of parameter setting, makes the choice of the parameters not depend on the user's experience, and greatly improves the accuracy as well as efficiency of segmentation. © 2012 Springer-Verlag.
CITATION STYLE
Zhang, X., Tong, H., & Chen, X. (2012). Multi-scale segmentation algorithm parameters optimization based on evolutionary computation. In Communications in Computer and Information Science (Vol. 316 CCIS, pp. 347–358). https://doi.org/10.1007/978-3-642-34289-9_39
Mendeley helps you to discover research relevant for your work.