Subspace Outliers Detection by Signal Subspace Matching

We present a novel and computationally efficient solution to the problem of subspace outlier detection that does not assume knowledge of the number of outliers nor exact knowledge of the dimension of the inliers subspace. The solution is based on a powerful representation of the inliers subspace, referred to as soft projection, and on a novel goodness-of-fit metric, referred to as signal subspace matching (SSM). Experimental results, demonstrating the performance of the SSM solution, are included.