Multi-Label Image Categorization With Sparse Factor Representation

• Published in: IEEE transactions on image processing Vol. 23; no. 3; pp. 1028 - 1037
• Main Authors: Fuming Sun, Jinhui Tang, Haojie Li, Guo-Jun Qi, Huang, Thomas S.
• Format: Journal Article
• Language: English
• Published: New York, NY IEEE 01.03.2014; Institute of Electrical and Electronics Engineers; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Algorithms; Applied sciences; Artificial Intelligence; Classification; Correlation; Documentation - methods; Exact sciences and technology; Image categorization; Image Enhancement - methods; Image Interpretation, Computer-Assisted - methods; Image processing; Information, signal and communications theory; Labels; Linear programming; multilabel; Natural Language Processing; Pattern Recognition, Automated - methods; Representations; Reproducibility of Results; Semantics; Sensitivity and Specificity; Signal and communications theory; Signal processing; Signal representation. Spectral analysis; Signal, noise; sparse; Sparse matrices; Tasks; Telecommunications and information theory; Terminology as Topic; Training; Transaction processing; Vectors; Performance evaluation; Multilabeling; Automatic classification; Image processing; Parcimony analysis; Algorithm; Signal classification; Modeling; Multiple image; Learning; multilabel; Accuracy; sparse; Sparse representation; Image categorization; Open market; Categorization; Image classification
• ISSN: 1057-7149; 1941-0042; 1941-0042
• DOI: 10.1109/TIP.2014.2298978

The goal of multilabel classification is to reveal the underlying label correlations to boost the accuracy of classification tasks. Most of the existing multilabel classifiers attempt to exhaustively explore dependency between correlated labels. It increases the risk of involving unnecessary label dependencies, which are detrimental to classification performance. Actually, not all the label correlations are indispensable to multilabel model. Negligible or fragile label correlations cannot be generalized well to the testing data, especially if there exists label correlation discrepancy between training and testing sets. To minimize such negative effect in the multilabel model, we propose to learn a sparse structure of label dependency. The underlying philosophy is that as long as the multilabel dependency cannot be well explained, the principle of parsimony should be applied to the modeling process of the label correlations. The obtained sparse label dependency structure discards the outlying correlations between labels, which makes the learned model more generalizable to future samples. Experiments on real world data sets show the competitive results compared with existing algorithms.