Indexing and retrieval of visually similar images in the orthogonal polynomials transform domain

Authors:
R. Krishnamoorthy;J. Kalpana
Affiliations:
Anna University of Technology Tiruchirappalli, Tiruchirappalli, India;Bharathidasan Institute of Technology, India
Venue:
ICDEM'10 Proceedings of the Second international conference on Data Engineering and Management
Year:
2010

Citing 4
Cited 1

A continuous probabilistic framework for image matching

Computer Vision and Image Understanding
A Comparative Study of Performance Measures for Information Retrieval Systems

ITNG '06 Proceedings of the Third International Conference on Information Technology: New Generations
Transform coding of monochrome images with a statistical design of experiments approach to separate noise

Pattern Recognition Letters
DCT-Domain image retrieval via block-edge-patterns

ICIAR'06 Proceedings of the Third international conference on Image Analysis and Recognition - Volume Part I

Fast communication: Generalized adaptive Bayesian Relevance Feedback for image retrieval in the Orthogonal Polynomials Transform domain

Signal Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

A Bayesian architecture for annotating, categorizing and retrieving 3D models of homogenous images given their 2D view is presented. Although the superiority of bayesian retrieval in a generic database has been studied, its ability to discriminate visually similar images, similarity being in colour, texture or shape has not been much reported. In the current work, we have established that continuous probabilistic image modeling based on mixture of Gaussians together with KL similarity measure, shows remarkable performance. For training, the characteristic view of the images is used. The features extracted are the polynomials transform coefficients. The algorithms used are simple, computationally efficient and do not require any prior segmentation. The dependence of the performance of the proposed architecture on the number of transform subspaces and the number of Gaussian mixtures has been studied. A comparative study with Daubechies wavelet shows that this architecture performs well with a small number of dimensions of transform subspaces and also with a small number of mixture of Gaussians, in addition to being fast.