Ensemble Combination for Solving the Parameter Selection Problem in Image Segmentation
Unsupervised image segmentation is of essential relevance for many computer vision applications and remains a difficult task despite of decades of intensive research. In particular, the parameter selection problem has not received the due attention in the past. Researchers typically claim to have empirically fixed the parameter values or train in advance based on manual ground truth. These approaches are not optimal and lack an adaptive behavior in dealing with a particular image. In this work we adopt the ensemble combination principle to solve the parameter selection problem in image segmentation. It explores the parameter space without the need of ground truth. The experimental results including a comparison with ground truth based training demonstrate the effectiveness of our framework.
KeywordsImage segmentation parameter selection ensemble combination concept multiple segmentation combination
- 2.Cingue, L., Cucciara, R., Levialdi, S., Martinez, S., Pignalberi, G.: Optimal range segmentation parameters through genetic algorithms. In: Proc. of 15th Int. Conf. on Pattern Recognition, Barcelona, vol. 1, pp. 474–477 (2000)Google Scholar
- 3.Felzenszwalb, P., Huttenlocher, D.: Efficient graph-based image segmentation. Int. Journal of Computer Vision, 167–181 (2004)Google Scholar
- 6.Martin, D., Fowlkes, C., Tal, D., Malik, J.: A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In: Proc. 8th Int. Conf. Computer Vision, vol. 2, pp. 416–423 (2001)Google Scholar
- 8.Min, J., Powell, M., Bowyer, K.W.: Automated performance evaluation of range image segmentation algorithms. IEEE Trans. on SMC – Part B 34(1), 263–271 (2004)Google Scholar
- 11.Wattuya, P., Jiang, X., Praßni, S., Rothaus, K.: A random walker based approach to combining multiple segmentations. In: Proc. of 19th Int. Conf. on Pattern Recognition, Florida (2008)Google Scholar