Analysis of learning using segmentation models
A method for change detection using a combination of view-based and model-based representations is presented. Coarse geometric models are enhanced with appearance-based information using a small number of training images, resulting in a hybrid representation that maps local appearance characteristics onto object subparts to provide model-based appearance prediction. The learning behavior of the change detection system is studied as a function of the number and characteristics of the images in the training set. Results are presented showing the level of improvement of change detection performance as images are added to the training set, in comparison with using a purely geometric approach.
Unable to display preview. Download preview PDF.
- 1.M. Bejanin, A. Huertas, G. Medioni, and R. Nevatia. Model validation for change detection. Proceedings of the ARPA Image Understanding Workshop, November 1994.Google Scholar
- 2.M. Costa and L. Shapiro. Scene analysis using appearance-based models and relational indexing. Proceedings of the International Symposium on Computer Vision, November 1995.Google Scholar
- 3.A. Hoogs and R. Bajcsy. Segmentation modeling. Proceedings of IAPR Conference on Computer Analysis of Images and Patterns, September 1995.Google Scholar
- 4.A. Hoogs and R. Bajcsy. Model-based learning of segmentations. Proceedings of ICPR, Vol. 4, August 1996.Google Scholar
- 5.H. Murase and S. Nayar. Learning object models from appearance. Proceedings of AAAI. Recognition, July 1993.Google Scholar
- 6.A. Pope and D. Lowe. Learning object recognition models from images. Proceedings of the 4th ICCV, May 1993.Google Scholar
- 7.M. Wheeler and K. lkeuchi. Sensor modeling, probabilistic hypothesis generation, and robust localization for object recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 17(3), March 1995.Google Scholar