Label Dependency Classifier using Multi Feature Graph Convolution Networks for Automatic Image Annotation
Vikas R. Palekar, Satish Kumar L · 2021 International Conference on Computational Performance Evaluation (ComPE) · 2021
Large scale image repository superintendence and retrieval of intelligent images is crucial issue due to abundant growth of images on digital platform. Automatic Image Annotation (AIA) certainly helps in this regards with huge influential development in machine learning technique. However gradual evolution methods developed by many researchers still demand the manageability and efficiency in large image dataset. High level semantic concept in correlation with low level image feature can be useful for huge dataset image retrieval. Additionally traditional approach of assigning labels manually for growing image dataset is practically not suitable as it is time-intensive and costly. Manual assignment of labels may lead to Error-prone and difficult to get quality labels. Only limited number training images with appropriate labels are available. Description associated with images on digital platform can help up to some extent to uplift the performance of AIA. We propose multiple label identification framework for AIA. In this framework we have used the joint optimization of multi label graph, Modality feature correlation and latent semantics. Semi supervised machine learning approach is used for Image annotation as image description associated with image can suggest good labels and results can be improved. Capacious experimentation is done on NUS-WIDE and MIR Flickr dataset. Results indicate the effective improvement in AIA .