Depth-based color stereo images retrieval using joint multivariate statistical models

• Published in: Signal processing. Image communication Vol. 76; pp. 272 - 282
• Main Authors: Ghodhbani, E., Kaaniche, M., Benazza-Benyahia, A.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.08.2019; Elsevier BV; Elsevier
• Subjects: Color; Color stereo images; Content-based retrieval; Depth map; Engineering Sciences; Gaussian copula; Gaussian process; Generalized Gaussian distribution; Image management; Image retrieval; Indexing; Multivariate analysis; Multivariate modeling; Parameter estimation; Statistical models; Three dimensional models; Wavelet transform; Wavelet transform; Depth map; Multivariate modeling; Content-based retrieval; Gaussian copula; Generalized Gaussian distribution; Color stereo images
• ISSN: 0923-5965; 1879-2677
• DOI: 10.1016/j.image.2019.05.008

The growing interest in using the three dimensional information in various application fields has led to the generation of huge color stereo image databases. As a result, it becomes necessary to design efficient content-based image retrieval systems well adapted to the indexing of such large databases. To this end, we propose in this paper different statistical-based retrieval approaches where the associated estimated model parameters are considered as a feature vector in the indexing process. More precisely, the Gaussian copula based multivariate Generalized Gaussian model will be used to capture the different correlations existing in color stereo images. While the first strategy aims at exploiting the cross-view as well as the cross-color channel redundancies, the second one resorts to a more general joint statistical model exploiting the correlation between the texture and depth information. Experimental results, performed on various datasets, confirm the benefits that can be drawn from the proposed approaches. •Efficient techniques for color stereo image retrieval are proposed.•The developed methods are based on Gaussian-copula based multivariate statistical models.•The models aim to exploit the cross-view and channel dependencies.•Joint statistical modeling of texture and depth information is also investigated.