This paper proposes an association method between image contents and their impression words for content-based image retrieval. Currently most of the image retrieval systems are based on ambiguous index terms assigned by the indexer or contents creator. Important issues are to improve the appropriateness and to reduce the effort of assigning the index terms. In this paper, we propose an analysis method of the relations between the content and the impression words (e.g. " fresh"," natural"," pop") given by the professional photographers, and applying them to automatic image classification system based on visual impression. In our method, according to visual impressions, a user classifies training examples into some groups labeled with impression words (impression group). Commercial photographs giving us similar impressions have typical arrangement patterns of objects. By applying clustering method based on minimum description length principle, we estimated a composition common to the photographs giving us similar impressions. To each region of the estimated composition, by constructing classifiers for the impression groups using SVM, and integrating the outputs of the classifiers using region importance factor, we have achieved 70 to 80% recognition ratio, 10% better than the method without composition analysis.
[1]
Bernhard Schölkopf,et al.
Estimating the Support of a High-Dimensional Distribution
,
2001,
Neural Computation.
[2]
K. Wakimoto,et al.
Efficient and Effective Querying by Image Content
,
1994
.
[3]
R. Haber,et al.
Visual Perception
,
2018,
Encyclopedia of Database Systems.
[4]
Amarnath Gupta,et al.
Visual information retrieval
,
1997,
CACM.
[5]
Jorma Rissanen,et al.
Fisher information and stochastic complexity
,
1996,
IEEE Trans. Inf. Theory.
[6]
Toshikazu Kato,et al.
Intelligent visual interaction with image database systems-toward the multimedia personal interface
,
1991
.
[7]
Shih-Fu Chang,et al.
VisualSEEk: a fully automated content-based image query system
,
1997,
MULTIMEDIA '96.
[8]
Vladimir Vapnik,et al.
Statistical learning theory
,
1998
.
[9]
Jitendra Malik,et al.
Blobworld: Image Segmentation Using Expectation-Maximization and Its Application to Image Querying
,
2002,
IEEE Trans. Pattern Anal. Mach. Intell..
[10]
James Ze Wang,et al.
SIMPLIcity: Semantics-Sensitive Integrated Matching for Picture LIbraries
,
2000,
IEEE Trans. Pattern Anal. Mach. Intell..