earticle

영어

In this paper, we propose new image retrieval system using global and local image features and the Dirichlet process Gaussian mixture model (DPGMM). First, we considered global and local image features. Global features are color histogram, wavelet coefficients, and Fourier descriptors extracted from a given image. Local features are defined as both the histogram features of pixel values extracted from square image patches and descriptors such as SIFT extracted from the salient region having center points as affine invariant detectors. Second, we have modeled an observed image as the DPGMM, and we have investigated the variational Bayesian inference method which can be used to estimate the parameters of DPGMM. And then we have extracted two types of feature vectors from an estimated DPGMM to represent a given image. Third, the image retrieval is conducted by matching two types of feature vectors such as the probability density feature and the feature signature vector generated by DPGMM based on two kinds of distance measures. Finally, we have carried out experiments on two kinds of real images datasets in order to compare the performance between the proposed method and the existing methods.