1: \begin{abstract}

2: Principal Component Analysis (PCA) is a fundamental method for

3: estimating a linear subspace approximation to high-dimensional

4: data. Many algorithms exist in literature to achieve a statistically

5: robust version of PCA called RPCA.  In this paper, we present a

6: geometric framework for computing the principal linear subspaces in

7: both situations that amounts to computing the intrinsic average on the

8: space of all subspaces (the Grassmann manifold).  Points on this

9: manifold are defined as the subspaces spanned by $K$-tuples of

10: observations.

11: % The Grassmannian in this

12: %case is defined using linear subspaces formed from collection of 'K'

13: %image frames off of incoming input data.

14: We show that the intrinsic Grassmann average of these subspaces

15: coincide with the principal components of the observations when they

16: are drawn from a Gaussian distribution.

17: %leads to recovery of the full set of principal

18: %components.

19: Similar results are also shown to hold for the RPCA. Further, we

20: propose an efficient online algorithm to do subspace averaging which

21: is of linear complexity in terms of number of samples and has a linear

22: convergence rate.  When the data has outliers, our proposed online

23: robust subspace averaging algorithm shows significant performance

24: (accuracy and computation time) gain over a recently published RPCA

25: methods with publicly accessible code.  We have demonstrated

26: competitive performance of our proposed online subspace algorithm

27: method on one synthetic and two real data sets. Experimental results

28: depicting stability of our proposed method are also

29: presented. Furthermore, on two real outlier corrupted datasets, we

30: present comparison experiments showing lower reconstruction error

31: using our online RPCA algorithm.  In terms of reconstruction error and

32: time required, both our algorithms outperform the competition.

33: \end{abstract}