Love Thy Neighbors: Image Annotation by Exploiting Image Metadata

TL;DR

This paper introduces a nonparametric approach that leverages image metadata to improve multilabel image annotation by utilizing neighborhoods of related images, outperforming existing methods especially with new metadata types.

Contribution

The paper presents a novel nonparametric method that uses image metadata to form neighborhoods, enhancing multilabel annotation performance and generalization to unseen metadata.

Findings

Outperforms state-of-the-art methods on NUS-WIDE dataset

Effective even with new, unseen metadata types

Utilizes a deep neural network to blend visual and neighborhood information

Abstract

Some images that are difficult to recognize on their own may become more clear in the context of a neighborhood of related images with similar social-network metadata. We build on this intuition to improve multilabel image annotation. Our model uses image metadata nonparametrically to generate neighborhoods of related images using Jaccard similarities, then uses a deep neural network to blend visual information from the image and its neighbors. Prior work typically models image metadata parametrically, in contrast, our nonparametric treatment allows our model to perform well even when the vocabulary of metadata changes between training and testing. We perform comprehensive experiments on the NUS-WIDE dataset, where we show that our model outperforms state-of-the-art methods for multilabel image annotation even when our model is forced to generalize to new types of metadata.