Local Best-Fit (LBF) Algorithm

---

Brief Introduction
LBF is a very simple geometric method for hybrid linear modeling. It is based on selecting a set of local best fit flats, where the size of the local neighborhoods is determined automatically by the Jones' beta_2 numbers. It can then select the global K flats by many of the existing strategies. In the current code we used an L_1 minimization procedure, but other successful procedures will be provided online soon. If the number of clusters K is unknown, we then apply the SOD algorithm provided below.

Paper
Randomized Hybrid Linear Modeling by Local Best-fit Flats

Matlab Code

• Local Best-fit Flats
• SOD (Second Order Difference) Algorithm
• Find the number of cluster by GPCA
  

Codes for other algorithms in the paper

• SCC and K-flats(by Guangliang Chen)
  

Supplemental Data

1. Artificial data: generated by using 'generate_samples.m' which is contained in the GPCA-voting folder
   
2. Real data: Motion Segmentation (http://www.vision.jhu.edu/data/hopkins155/)

Contact

• Gilad Lerman, lerman@umn.edu
• Teng Zhang, zhang620@umn.edu
• Arthur Szlam, aszlam@courant.nyu.edu
• Yi Wang, wangx857@umn.edu

Acknowledgement

• The research described here was supported by NSF grants DMS-0612608, DMS-0811203 and DMS-0915064.