Content-based Selection of Methods for Image Segmentation

G. Peters and M. Petke (Germany)


Image Segmentation, Neural Networks, Edutainment, Dig ital Art


Many different methods for image segmentation have been developed. Each of them usually has its advantages for one single class of images only. An unsolved problem in the field of image segmentation, however, consists in the fact that there does not exist a single approach which can be applied to several classes of images with the same success. As an idea for a solution of this problem we propose an au tonomous selection of a segmentation method by a neural network depending on the specific content of the image to be segmented. The neural network is trained with statistical image measures of classical modern paintings and is able afterwards to select the most appropriate of four standard methods for image segmentation for unknown test images. We evaluate our approach in the field of edutainment by an automatization of Ursus Wehrli’s idea of ”Cleaning up artworks”.

Important Links:

Go Back