SIMAR: Segmentation and Fusion of Images with Multimodality at Different Resolutions
Often images are obtained from different sources (multimodality) and must be combined in order to classify the data into particular segments. For example we may have images of the ground acquired by various Unmanned Ariel Vehicles (UAV) engaged in a search and rescue mission. Such images will be at different resolutions if UAVs are operating at different altitudes and we can also fuse such images with those obtained from satellites.
We propose to extend previous work on a model-based approach for intensity-based segmentation of images acquired from multiple independent modalities. Pixel intensity within a modality image is represented by a Gaussian distribution mixture model where the components of the mixture correspond to different segments. Segmentation classes are then determined by maximising a posteriori probability contributed from all multimodal images. Experimental results show that the method exploits and fuses partially redundant and complementary information of multimodal images. Segmentation can thus be more precise than where using single-modality images.
This PhD will generalise our previous approach to segmentation of multi-resolution images where some of the multi-modal images are more detailed than others. We may also use the same method to incorporate contextual data, using Markov random fields to place neighbouring pixels into the same segment.
The PhD will be built on previous work within with the EPSRC funded SUAAVE project which involved researchers from University College London, the University of Oxford and the University of Ulster. SUAAVE focused on the creation and control of swarms of helicopter UAVs (unmanned aerial vehicles) that operate autonomously, collaborate to sense the environment, and report their findings to a base station on the ground. Search and rescue is the main focus although many other applications are envisaged. Various images and other data, obtained by on-board cameras and other sensors, are available on which to base a decision as to how the UAV might locate a missing person and also where a UAV might gracefully land, if required. Such data are typically multimodal and multi-resolution so may therefore be analysed using the segmentation algorithms that will be developed in this project.
Personnel Involved
First Supervisor: McClean, SI Prof
Second Supervisor: Scotney, BW Prof
Third Supervisor: Morrow, PJ Dr
Collaboration: This project does not involve collaboration with another establishment
Synopsis:
We propose to extend previous work on a model-based approach for intensity-based segmentation of images acquired from multiple independent modalities. Pixel intensity within a modality image is represented by a Gaussian distribution mixture model where the components of the mixture correspond to different segments. Segmentation classes are then determined by maximising a posteriori probability contributed from all multimodal images. Experimental results show that the method exploits and fuses partially redundant and complementary information of multimodal images. Segmentation can thus be more precise than where using single-modality images.
single-modality images.