Automatic Image Annotation Using Transfer Learning on Convolutional Neural Networks
Rasmus Ek · Doria (University of Helsinki) · 2018
Categorizing images in order to create image tags for easy searching is often a manual, and thus time-consuming, process. By utilizing a convolutional neural network to perform image classification, one can automate this categorization process, by using the detected image classes to automatically generate basic search tags. These tags can then be used to facilitate searching and linking of information, making the images easier to find, and providing links to other information, such as related images and explanatory texts. Convolutional neural networks have in recent years shown great promise in solving image classification problems, as demonstrated in various image recognition challenges and competitions. In this paper, the image classification capabilities of convolutional neural networks are demonstrated by creating, training, and using an Inception-based convolutional neural network to classify a set of images, providing categories that could be used for image annotation. In addition, a few alternative solutions for automatic image annotation are presented and discussed, to provide points of comparison for the created solution.