Image and medical annotations using non-homogeneous 2D ruler learning models. (February 2016)

• Record Type: Journal Article
• Authors: Hong, Lixia; Jin, Qingyue; Li, Xusheng; Huang, Yizhen
• Is Part Of: Computers & electrical engineering. Volume 50(2016)
• Page Start: 102
• Page End: 110
• Publication Date: 2016-02
• Subjects: Semidefinite programming -- Image annotation -- Medical annotation -- Prediction model; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2016.01.011
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: A metric learning model directly calibrating measured numerics of feature data. A 2D semi-metric model by relaxing one degree of freedom. Simple and intuitively natural models solvable by convex quadratic programming. Abstract: We propose a metric learning model called "non-homogeneous 2D rulers", in which the measured numerics of the observed feature data are directly calibrated on non-homogeneous 2D rulers. It is actually the definition of variables relaxed by one degree of freedom, which constitutes a 2D semi-metric model. Our proposed models are intuitively natural, and they are applied to solve various types of annotation problems, and particularly the recurrent spontaneous abortion prediction medical annotation problem. Experiments on the LabelMe, UIUC-sports, TRECVID 2005, MSRC, Barcelona, and our own collected clinical datasets show that, even with simple kNN, our models are competitive among the state of the art. Graphical abstract: