Research On Methods For Multiresolution Segmentation Of Infrared Images

Infrared imaging guidance technology is one of the important developmentdirections for the modern precision-guided weapons. And the target detection andrecognition technology under the condition of the infrared imaging is indispensable tothe infrared imaging guidance system. As the foundation of object recognition andtracking, the quality of infrared image segmentation is critical for improving theperformance of infrared pre-warning and guidance systems.The main task of this thesis is to research the multi-resolution segmentationalgorithms. Firstly, in this paper, the subjective evaluation criteria and commonobjective evaluation criteria for image segmentation is introduced. Since an individualevaluation criterion cannot accurately evaluate the performance of segmentation, agoodness measure which synthesizes five character indexes to reflect the performance isproposed. Experiment results show that the new measure can evaluate segmentationalgorithms effectively. Secondly, the thesis reviews the traditional thresholdsegmentation and area-growth segmentation applying in infrared segmentation,experiments on these algorithms and provides clues to explore other methods. Thirdly,as the traditional method of infrared image segmentations make no use of priorknowledge, the paper introduces a single resolution image segmentation based on MRFmodel in the Bayesian framework, experiments show that the algorithm can makesegmentation successfully. Finally, since the MRF model under the single resolutioncannot depict image features successfully, build MRF models in wavelet domainmulti-resolution image. Through comparative analysis of two separate experimentalmethods, we know that MRF model-based multi-resolution segmentation can captureimages in different resolution structural information, and reduce over-segmentationphenomenon, get better segmentation results.