Multi-Label Classification for Images with Labels for Image Annotation

Abstract

Images and videos are increasing due to advancement in digital technologies. Annotate the given image for efficient image retrieval and processing it is required identifying a set of objects present in each image. Manual text-based image annotation is very time consuming and expensive and it becomes infeasible with such exponential increase in visual data. Multi-label classification problemgeneralizes the standard multiclass classification by allowing each instance to be simultaneously assigned into multiple label categories. A key challenge for multi-label classification is label sparsity that is many labels lacks sufficient training instances for building efficient classifiers. Hence, exploiting label dependency can significantly boost classification performance. Most of multi-label method uses binary decomposition of multi-label datasets but uses the same features for training the classifiers which may contain redundant features.