High-order nonnegative blind source separation based on edge features

• Published in: Signal, image and video processing Vol. 17; no. 8; pp. 4163 - 4170
• Main Authors: Zhao, Mingzhan, Zheng, Weipeng, Lv, Yingli, Du, Chunmei, Wang, Zhiliang, Xu, Xiaojun
• Format: Journal Article
• Language: English
• Published: London Springer London 01-11-2023; Springer Nature B.V
• Subjects: Algorithms; Clustering; Complexity; Computer Imaging; Computer Science; Image Processing and Computer Vision; Multimedia Information Systems; Original Paper; Pattern Recognition and Graphics; Signal,Image and Speech Processing; Vision; NBSS; Edge feature; Projection; Slope distribution; Density distribution
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-023-02648-2

The nonnegative blind source separation (NBSS) algorithm based on minimum volume simplex (MVS) criterion is excessively dependent on the shape of the mixture scatterplot, resulting in the situation in which the MVS-based algorithm may have no solution. In this paper, we propose a new noiseless NBSS model and introduce edge features into high-order determined NBSS. The edge feature-based NBSS model can compensate for the shortcomings of the MVS-based algorithm. The twice projections (TP) algorithm is designed to replace the existing clustering and regression algorithms; moreover, TP prevents the aliasing phenomenon caused by direct projection and effectively reduces the time complexity of dimension reduction. We search for the coordinates of the density maximum points in 2-D space, and gradually merge them into high-dimensional coordinates, which greatly reduces the complexity of the algorithm. Furthermore, we only make boundedness and nonnegativity assumptions about the source, which makes the algorithm more widely applicable.