Multi-modal Image Fusion and Its Applications

Image fusion integrates different modality images to provide comprehensive information of the image content, increasing interpretation capabilities and producing more reliable results. There are several advantages of combining multi-modal images, including improving geometric corrections, complementing data for improved classification, and enhancing features for analysis…etc.

This thesis develops the image fusion idea in the context of two domains: material microscopy and biomedical imaging. The proposed methods include image modeling, image indexing, image segmentation, and image registration. The common theme behind all proposed methods is the use of complementary information from multi-modal images to achieve better registration, feature extraction, and detection performances.

In material microscopy, we propose an anomaly-driven image fusion framework to perform the task of material microscopy image analysis and anomaly detection. This framework is based on a probabilistic model that enables us to index, process and characterize the data with systematic and well-developed statistical tools. In biomedical imaging, we focus on the multi-modal registration problem for functional MRI (fMRI) brain images which improves the performance of brain activation detection.