Progressive Deep Non-Negative Matrix Factorization Architecture with Graph Convolution-based Basis Image Reorganization

• Published in: Pattern recognition Vol. 132; p. 108984
• Main Authors: Zhao, Yang, Deng, Furong, Pei, Jihong, Yang, Xuan
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2022
• Subjects: Basis image factorization; Basis image reconstruction; Deep non-negative matrix factorization; Face recognition; Graph convolution; Basis image reconstruction; Basis image factorization; Graph convolution; Face recognition; Deep non-negative matrix factorization
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2022.108984

•A novel progressive deep non-negative matrix factorization (PDNMF) architecture is proposed. Different from the existing deep NMF-based methods that continuously factorizes the basis images, the PDNMF progressively decomposes the basis images in a cyclic form of “factorization-reconstruction-factorization”.•An idea to describe the expressiveness of basis images and a method for semantic reconstruction of basis images are proposed based on graph convolution. The basis image reconstruction based on this idea will not severely destroy the expressive ability of each basis image. Meanwhile, the robustness of the basis image is enhanced to support its deeper factorization.•The basis image reconstructed based on proposed attribute similarity graph (ASG) can withstand deeper factorization.•Deeper factorization result can be obtained by PDNMF than existing NMF-based methods.•The recognition accuracy of PDNMF shows an overall upward trend with the increase of the number of layers. Deep non-negative matrix factorization is committed to using multi-layer structure to extract underlying parts-based representation. However, the basis images obtained by continuous depth factorization is too sparse, resulting in too fragmented parts reflected by the basis image. This makes the number of factorization layers limited and the underlying local feature representation is inaccurate. Therefore, we propose a novel progressive deep non-negative matrix factorization (PDNMF) architecture that adds a basis image reconstruction step to the successive basis image factorization steps. This helps the basis image in depth factorization to maintain better robustness of feature representation. In the reconstruction step, the attribute similarity graph (ASG) is constructed to describe the semantic expression ability of each basis image. With the help of the ASG, the basis image enhances its own semantic integrity through graph convolution without drastically destroying its representation. The evaluation in image recognition shows that the recognition accuracy of the proposed PDNMF improves with the increase of layers. Our method outperforms the state-of-the-art deep factorization methods in image recognition.