Outliers Detection by Signal Subspace Matching

We present a novel solution to the problem of subspace outlier detection that does not assume prior knowledge of the number of outliers nor the dimension of the inliers subspace. The solution is based on the recently introduced notion of soft projection for capturing the inliers subspace, and on the recently introduced signal subspace matching (SSM) metric for measuring the distance between the given vectors and the inliers subspace. The solution handles both unstructured and structured outliers and a relatively large ratio of outliers to inliers. Experimental results, demonstrating the performance of the SSM solution, are included.