Robust and discrete matrix factorization hashing for cross-modal retrieval. (February 2022)

• Record Type: Journal Article
• Authors: Zhang, Donglin; Wu, Xiao-Jun
• Is Part Of: Pattern recognition. Volume 122(2022)
• Publication Date: 2022-02
• Subjects: Cross-modal retrieval -- Hashing -- Autoencoder -- Discrete optimization,; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.108343
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We present a novel two-step hashing algorithm (RDMH) for cross-media retrieval. We develop a discrete matrix factorization scheme, the binary codes can be learned directly. The quantization errors caused by the rounding or relaxation scheme can be avoided. We utilize l 2, 1 -norm in the proposed framework to enhance the robustness of our method, which makes the proposed method robust to noises and outliers. We propose a novel autoencoder strategy to learn the hash functions. More valuable information can be preserved, making the hash functions more powerful. Abstract: Hashing based methods have gained great success for cross-modal similarity search, due to its fast query speed and low storage cost. However, there are some challenging problems that need to be further solved: 1) Many approaches are sensitive to noises and outliers, because ℓ 2 norm is utilized in the objective function, the error may be amplified. 2) Most existing methods take relaxation or rounding scheme to generate binary codes, causing a large quantization loss. 3) Many supervised cross-media algorithms usually take a large n × n matrix to preserve the similarity relationship, leading to large calculation and making them unscalable. To mitigate these challenges, we develop a novel cross-media search algorithm, i.e., robust and discrete matrix factorization hashing, dubbed RDMH. The method takes a two-step strategy. In the first phase, the ℓ 2, 1 norm is utilized to improve the robustness, which makes our model not sensitive to noises and outliers. We can learn the hash codes directly by the proposed discrete optimization method instead of relaxation scheme, avoiding the large quantization loss. Moreover, RDMH correlates the hash codes and semantic labels directly instead of manipulating the large similarity matrix. In the second phase, we propose an autoencoder strategy to learn the hash functions, more valuable information can be preserved and making the hash functions more powerful. Comprehensive experiments on several databases demonstrate the superior performance and efficacy of the developed RDMH.