Dear Guohua Lv,

I think thar the following paper  http://pages.cs.wisc.edu/~lizhang/vision-reading/2008s/BeyondRANSAC.pdf may be of interest to you. The authors propose a robust estimator, called the projection based M-estimator which is compared with RANSAC and MSAC. In particular, they say that their technique gives better inlier/outlier dichotomies and better parameter estimates than RANSAC.

Regards

Dear Marco Fiorucci,

Thanks for your suggestion. I will look into the paper and try to do experiments on this method.

Regards

Yes M-estimator works well. I recommend it.

You may also want to take a look to this paper: https://www.inf.ethz.ch/personal/pomarc/pubs/RaguramECCV08.pdf

It gives an overview on more recent approaches.

Thanks to Sebastian Zambanini. I will look into the paper you suggested.

Hi Guohua, 

I found this algorithm called  Least Median Squares algorithm (LSA). Its use is described in the first paper below and the second is its different variations. I hope you find them useful.

https://www.nuffield.ox.ac.uk/economics/papers/.../OutlierDetectionAlgorithms.pdf

https://arxiv.org/pdf/1304.1250

Best!

Thanks to Folasade Isinkaye. I will go through the two papers you suggested and do experiments if possible. Regards.

Dear Ghuohua,

There is also this work based on J-linkage clustering method.

"Copy-move forgery detection and localization by means of robust clustering with J-Linkage", Irene Amerini. Lamberto Ballan, RobertoCaldelli, Alberto Del Bimbo, Luca Del Tongo, Giuseppe Serra, Signal Processing: Image Communication: Volume 28, Issue 6, July 2013, Pages 659-669.

Bests!