Image retrieval on large-scale image databases

Abstract

Online image repositories such as Flickr contain hundreds of millions of images and are growing quickly. Along with that the needs for supporting indexing, searching and browsing is becoming more and more pressing. In this work we will employ the image content as a source of information to retrieve images. We study the representation of images by Latent Dirichlet Allocation (LDA) models for content-based image retrieval. Image representations are learned in an unsupervised fashion, and each image is modeled as the mixture of topics/object parts depicted in the image. This allows us to put images into subspaces for higher-level reasoning which in turn can be used to find similar images. Different similarity measures based on the described image representation are studied. The presented approach is evaluated on a real world image database consisting of more than 246,000 images and compared to image models based on probabilistic Latent Semantic Analysis (pLSA). Results show the suitability of the approach for large-scale databases. Finally we incorporate active learning with user relevance feedback in our framework, which further boosts the retrieval performance.

Multilayer pLSA for multimodal image retrieval

“…Topic models have been used in several previous works in order to derive a low dimensional image description suitable for large-scale image retrieval, for example [18] used probabilistic Latent Semantic Analysis (pLSA) [13] based models, [15] applied Latent Dirichlet Allocation (LDA) [5] to derive a topic representation and [9] adopted the Correlated Topic Model (CTM) [3]. However all of the previous In [1], [4] and [20] the authors propose topic models to model annotated image databases.…”

Accurate Image Search Using the Contextual Dissimilarity Measure

“…Comparison measures that do no satisfy the properties of a distance have been used in information retrieval. For instance, the image search system of [16] explores the use of the Shannon-Jenson divergence and a metric derived from a LDA model.…”

Filtering adult image content with topic models

“…Based on our prior work [7] we compute two different local feature descriptors from each local region centered at (x, y):…”

“…In this paper we analyze how well recent concepts from image classification in general can be exploited for filtering adult content. Recently very successful approaches to image classification use topic models on visual words derived from salient descriptors of local image patches [12,1,10,7]. The bestknown topic model is the probabilistic Latent Semantic Analysis (pLSA) [8].…”