In this paper, we propose a framework for image classification tasks, named MIC, that takes as input multi-view images, such as RGB-T images for surveillance purposes. We combine auto-encoder and generative adversarial network architectures to ensure the multi-view embedding in a common latent space. Then, the resulting features are fed to the classification stage. The proposed framework is able to, all at once, train the multi-view embedding model to find a shared latent representation for the different views, perform data imputation (generate the missing views) and ensure the classification task by predicting the labels. Experiments on the MNIST dataset with a panoply of classifiers and several missingness ratios show the effectiveness of our solution.
Gianmarco Aversano, Mahmoud Jarraya, Maher Marwani, Ichraf Lahouli and Sabri Skhiri, MIC: Multi-view Image Classifier using Generative Adversarial Networks for Missing Data Imputation, Proc. of the 18th IEEE International Multi-conference on Systems, Signals and Devices, 2021
Click here to access the paper.