Kernel-based distance metric learning for content-based image retrieval

Abstract

For a specific set of features chosen for representing images, the performance of a content-based image retrieval (CBIR) system depends critically on the similarity or dissimilarity measure used. Instead of manually choosing a distance function in advance, a more promising approach is to learn a good distance function from data automatically. In this paper, we propose a kernel approach to improve the retrieval performance of CBIR systems by learning a distance metric based on pairwise constraints between images as supervisory information. Unlike most existing metric learning methods which learn a Mahalanobis metric corresponding to performing linear transformation in the original image space, we define the transformation in * Corresponding author: Dr. Dit-Yan Yeung, dyyeung@cs.ust.hk, +852-2358-1477 1 This is the Pre-Published Version the kernel-induced feature space which is nonlinearly related to the image space.Experiments performed on two real-world image databases show that our method not only improves the retrieval performance of Euclidean distance without distance learning, but it also outperforms other distance learning methods significantly due to its higher flexibility in metric learning.

A Kernel Approach for Semisupervised Metric Learning

“…As another direction, we will incorporate dissimilarity constraints into the methods to further improve the metric learning performance. Moreover, we will explore the application of the proposed methods to other real-world problems such as content-based image retrieval [7], [8], [9].…”

Content-Based Retrieval of Focal Liver Lesions Using Bag-of-Visual-Words Representations of Single- and Multiphase Contrast-Enhanced CT Images

“…DML methods can be used to find a linear transformation that projects the image features to a new meaningful feature space to reduce this semantic gap. Previous work showed that appropriately designed distance metrics could improve CBIR performance compared with Euclidean distance [27]. For the BoW model, the semantic meaning of visual words is ambiguous; thus, the retrieval performance can be improved by embedding the semantic information to BoW representation by supervised DML.…”

Similar Speaker Selection Technique Based on Distance Metric Learning Using Highly Correlated Acoustic Features with Perceptual Voice Quality Similarity

“…Multiple regression analysis is widely used since it can weight each feature, but "distance metric learning [14]" (DML) is more effective since it can take side information into account. Many studies on DML have demonstrated its usefulness in applications such as image retrieval [15], music retrieval [16], and sentence retrieval [17]. This technique can realize speaker selection if the side information is set properly.…”

“…In addition, DML has also been used for feature space transformation in a number of studies. For instance, [15] used transformation of the original image space for image retrieval. In this paper, since the perceptual voice quality similarity is used as the side information, DML can be considered to be transformation from acoustic feature space to perceptual voice quality similarity space.…”