Multi-Modal Images

Multi-modal images are produced by merging images of different modalities in order to produce digital images that contain important information for certain tasks. There are numerous domains where multi-modal images are essential for performing particular tasks. In remote sensing, images captures through synthetic aperture radar (SAR) systems have the advantage of operating independently of atmospheric conditions such as clouds and rain.

In medical applications, positron emission tomography (PET) produce images that contain crucial information for physicians. Also, hyperspectral imaging techniques have received significant amounts of attention recently, as they can provide images with important depth information in skin cancer diagnostics. Processing multi-modal images using computer vision systems can be a challenging task, as many standard image processing techniques are not applicable. The UiT Machine Learning Group has many years of experience with handling multi-modal images and has developed a number of techniques that are tailored to handle the difficulties that may arise when dealing with such imagery.

Highlighted Publications

1. Unsupervised Image Regression for Heterogeneous Change Detection (IEEE Transactions on Geoscience and Remote Sensing 2019).

Luigi T. Luppino, Filippo M. Bianchi, Gabriele Moser and Stian N. Anfinsen

2. Urban land cover classification with missing data using deep convolutional neural networks (2017 IEEE International Geoscience and Remote Sensing Symposium)

Michael Kampffmeyer, Arnt-Børre Salberg and Robert Jenssen