Talk:RANSAC

From Wikipedia, the free encyclopedia

This article is within the scope of WikiProject Statistics, which collaborates to improve Wikipedia's coverage of statistics. If you would like to participate, please visit the project page.

WikiProject Echo has identified RANSAC as a foreign language featured article. You may be able to improve this article with information from the German language Wikipedia.

[edit] interpretation of output

Does the algorithm's output (assuming convergence, etc) have a well-defined interpretation in terms of some bayesian (or other) model of the data and outliers? I believe it does, but I don't know what it is. -- David W. Hogg 00:23, 1 June 2007 (UTC)

[edit] Iteration?

Does this algorithm iteratively converge on a result? As it is described, it sounds like it just repeatedly selects a random set. It seems like it could do better by selecting a random set, then adding extra points that fit well and excluding points that don't fit well until that converges. 155.212.242.34 19:19, 17 October 2007 (UTC)

About convergence. Your question on convergence is not easy to answer since it depends on what we mean by convergence. The algorithm iteratively seeks to improve its current solution, i.e, any time is selects a new solution it is a better solution than the previous one, which means that if it is allowed to continue its search indefinitely, it will "converge" to the global optimum. However, in practice it cannot iterate indefinitely, typically it iterates a fix number of times. This means that there is a small but non-zero probability that it does not find even a close-to-optimal solution. The algorithm is often characterized as "indeterministic" since it provides reasonable results only with a certain probability, and this probability increases with the number of iterations which are allowed. --KYN 11:01, 18 October 2007 (UTC)

About improving the algorithm. What you are proposing may or may not be a good idea, depending on the application. It is, however, not how the basic RANSAC algorithm works. On the other hand, there are plenty of extensions of the original algorithm which can be found in the literature, and your proposal could probably be one of them. --KYN 11:01, 18 October 2007 (UTC)

Thanks for the clarification. It sounds almost too simple to make much headway, but I guess it works. 155.212.242.34 16:08, 18 October 2007 (UTC)