High-order nonnegative blind source separation based on edge features

Abstract

The edge feature-based Nonnegative Blind Source Separation (NBSS) proposed by us can make up for the shortcomings of the Minimum Volume Simplex (MVS) based algorithm that limits the shape of the scatterplot. In this paper, we design a high-order NBSS algorithm based on edge features. The algorithm adopts the twice projections, which avoids the aliasing phenomenon caused by the direct projection and effectively reduces the time complexity of dimensionality reduction. We searched for the coordinates of the density maximum points in the 2-D space, and gradually merged them into high-dimensional coordinates, which greatly reduced the complexity of the algorithm. Furthermore, we only make Boundedness and Nonnegativity assumptions about the source, which makes the algorithm more widely applicable.