Conclusions
- Accuracy: the particular feature vector used in this implementation promotes clustered classes, thus increasing accuracy;
- Speed: when the number of classes to evaluate is small, EM is fast at separating the classes. The slow step is the evaluation of the feature vector for all the points in the image;
Future Work
An interesting idea to expand this work is to use multiple images (of the same scene) to obtain a better sample of the points.
References
- J. Cai, A. Goshtasby, Image and Vision Computing 18 (1999) 63-75
- M.J. Swain, D.B. Ballard, Intl. Journal of Computer Vision, 7:1 (1997) 11-32
- V.I. Pavlociv, R. Sharma, T.S Huang, IEEE Trans. on Patt. Anal. and Machine Intel., 19:7 (1997) 677-694