Accurate tag relevance prediction for image search

Abstract

Embodiments of the present invention provide an automated image tagging system that can predict a set of tags, along with relevance scores, that can be used for keyword-based image retrieval, image tag proposal, and image tag auto-completion based on user input. Initially, during training, a clustering technique is utilized to reduce cluster imbalance in the data that is input into a convolutional neural network (CNN) for training feature data. In embodiments, the clustering technique can also be utilized to compute data point similarity that can be utilized for tag propagation (to tag untagged images). During testing, a diversity based voting framework is utilized to overcome user tagging biases. In some embodiments, bigram re-weighting can down-weight a keyword that is likely to be part of a bigram based on a predicted tag set.

Description

technical field [0001] Various embodiments of the present application generally relate to the field of image search, and in particular relate to accurate tag correlation prediction for image search. Background technique [0002] Internet-based search engines have traditionally employed common image search techniques to locate digital image content on the World Wide Web. One of these well-known image search techniques can be classified as "text-based" image search. A traditional text-based image search may receive a text-based query that is used to search a database of keyword-tagged images to generate images each with one or more keyword tags that match the text-based query. The resulting image collection. These text-based searches primarily rely on the quality and level of detail of the keyword tags in the image databases on which the search is performed. These keyword tags are usually provided by an automated tagging system. [0003] Current labeling systems treat labe...

AI Technical Summary

Problems solved by technology

However, current clustering algorithms often result in unbalanced data, where most data points (e.g., images) are in the same cluster, leaving other clusters w

Images