Open AccessCross-Media Retrieval Based on Canonical Correlation Analysisand Decision Tree
Author(s) -
Hongye Duan,
Meng Wang,
Peiqing Zou
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1682/1/012044
Subject(s) - canonical correlation , computer science , subspace topology , data mining , feature (linguistics) , information retrieval , correlation , tree (set theory) , decision tree , artificial intelligence , pattern recognition (psychology) , mathematics , geometry , mathematical analysis , philosophy , linguistics
The retrieval of approximate data from massive multimedia data is the key of computer science research. With the explosive growth of data scale, data retrieval needs to face the overlapping massive multimodal data called “dimension disaster”, and the traditional retrieval methods appeared to be inefficient. The cross-media retrieval with canonical correlation analysis (CCA) is a kind of way to map different media features to the largest correlation isomorphism subspaceand then compare the similarity between cross-media data in this subspace. However CCA is a linear model and it is difficult to map the lower data to the higher semantic. This paper proposed a cross-media retrieval method based on CCA and Decision Tree to solve the problem. CCA was uesd to depict the correlation between image and text feature, Decision Tree was uesd toapproach to feedback and repeatedly adjust the correlation. Experiments on the Wikipedia text-image datasetsverified that the Tree-CM model can mine the complex correlation between cross-media data and has better performance than other state-of-the-art models.