Dealing with uncertainty and imprecision in image segmentation using belief function theory

• A new method to segment multi-source images using belief function.
• Dealing with uncertainty and imprecision separately.
• Neighbourhood information fusion to reduce uncertainty and to underline imprecision.
• Multi-modal information fusion to remove efficiently imprecision.
• Good performances obtained on both simulated images and medical images.

In imaging, physical phenomena and the acquisition system are responsible for noise and the partial volume effect, respectively, which affect the uncertainty and the imprecision. To address these different imperfections, we propose a method that is based on information fusion and that uses belief function theory for image segmentation in the presence of multiple image sources (multi-modal images). First, the method takes advantage of neighbourhood information from mono-modal images and information from an acquisition system to reduce uncertainty from noise and imprecision due to the partial volume effect. Then, it uses information that arises from each modality of the image to reduce the imprecision that is inherent in the nature of the images, to achieve a final segmentation. The results obtained on simulated images using various signal-to-noise ratios and medical images show its ability to segment correctly multi-modal images in the presence of noise and the partial volume effect.