Re-ranking of web image search results using a graph algorithm

Abstract

We propose a method to improve the results of image search engines on the Internet to satisfy users who desire

Latent visual context learning for web image applications

“…In literature, there are many works about visual re-ranking based on different schemes, such as clustering-based [15,21], classification-based [19] and the graph-based [8,[16][17][18]20,22,24]. Generally, an assumption is made that visually similar images should be ranked close to each other.…”

“…In all visual re-ranking methods, an essential problem is how to measure the visual similarity. Currently, the similarity is mainly estimated based on low-level visual features: global features [17,18,24], such as color moments and Gabor feature, and local features [8,20,22,24], such as scale invariant feature transform (SIFT) [2]. Global features work well for cases such as natural scene images, while local features do good job in rigid canonical object images.…”

Context-based image re-ranking for content-based image retrieval

“…They propose a relevance model which is a probabilistic model that evaluates the relevance of the HTML document linking to the image, and assigns a relevance probability, so that the ranking could consider the local text information such as caption as well as the global text information. On the other hand, several methods [5], [6], [7] based on visual information only are also proposed. Zitouni et al [5] present the similarities of all images in a graph structure, find the densest component that corresponds to the largest subset of most similar images and give higher priority to them.…”

“…On the other hand, several methods [5], [6], [7] based on visual information only are also proposed. Zitouni et al [5] present the similarities of all images in a graph structure, find the densest component that corresponds to the largest subset of most similar images and give higher priority to them. Their work is based on an assumption, results of text based systems include a subset of correct images, and this set is, in general, the largest one which has the most similar images compared to other possible subsets.…”

A hybrid relevant-diverse approach for image re-ranking with multiple features

“…They conducted experiments on product images, and the experimental results showed that VisualRank can improve the relevancy and diversity of image search results. In [36], the authors presented similarities of all images in a graph structure and found the densest component that corresponds to the largest set of most similar subset of images. Then, to re-rank the results, they gave higher priority to the images in the densest component and ranked the others based on their similarities to the images in the densest component.…”