Image tag refinement by regularized latent Dirichlet allocation

Abstract

Tagging is nowadays the most prevalent and practical way to make images searchable. However, in reality many manually-assigned tags are irrelevant to image content and hence are not reliable for applications. A lot of recent efforts have been conducted to refine image tags. In this paper, we propose to do tag refinement from the angle of topic modeling and present a novel graphical model, regularized Latent Dirichlet Allocation (rLDA). In the proposed approach, tag similarity and tag relevance are jointly estimated in an iterative manner, so that they can benefit from each other, and the multi-wise relationships among tags are explored. Moreover, both the statistics of tags and visual affinities of images in the corpus are explored to help topic modeling. We also analyze the superiority of our approach from the deep structure perspective. The experiments on tag ranking and image retrieval demonstrate the advantages of the proposed method.

Latent Dirichlet allocation (LDA) and topic modeling: models, applications, a survey

“…This model can estimates the latent topics for each document, with making use of other documents. They used a collective inference scheme to estimate the distribution of latent topics and applied a deep network structure to analyze the benefit of regularized LDA [45][46].…”

“…In other view, Most of the papers that were studied, had goals for this topic modeling, such as: Source code analysis [8 , 9, 22, 24-27] , Opinion and aspect Mining [18,[28][29][30][31][32][33][34][35][36], Event detection [37][38][39][40], Image classification [13,41], recommendation system [42][43][44][45][46][47][48] and emotion classification [49][50][51], etc. For example in recommendation system, in [44] proposed a personalized hashtag recommendation approach based LDA model that can discover latent topics in microblogs, called Hashtag-LDA and applied experiments on "UDI-TwitterCrawl-Aug2012-Tweets" as a real-world Twitter dataset.…”

Socializing the Semantic Gap

“…In particular, the authors extend the Latent Dirichlet Allocation model [Blei et al 2003] to force images with similar visual content to have similar topic distribution. According to their experiments [Wang et al 2014], however, the gain of such a regularization appears to be marginal compared to the standard Latent Dirichlet Allocation model. first find embedding vectors of training images and tags using the image-tag association matrix of S. The embedding vector of a test image is obtained by a convex combination of the embedding vectors of its neighbors retrieved in the original visual feature space.…”

“…Notice that the models can be tag specific or holistic for all tags. As an example of holistic modeling, a topic model approach is presented in Wang et al [2014] for tag refinement, where a hidden topic layer is introduced between images and tags. Consequently, the tag relevance function is implemented as the dot product between the topic vector of the image and the topic vector of the tag.…”

Unsupervised Greenhouse Tomato Plant Segmentation Based on Self-Adaptive Iterative Latent Dirichlet Allocation from Surveillance Camera

“…The latent Dirichlet allocation, LDA, first proposed by Blei, Ng et al [32], is one of the most popular topic generation models, which contains a three-level structure of word, topic, and document. Recently, it has developed rapidly and become distinguished in the field of image processing such as image recognition, classification, annotation, and so on [33,34]. Its fundamental conception, bag-of-words model (BoW) [35,36], was originally used in distinguishing hidden information in a large collection of corpus [37,38] and conversing the information of the pixels to non-ordered visual words.…”