Multispectral images based bridge detection method with RX detector

Bridge plays an important role in people's life. The automatic bridge detection has great value especially in disaster situation. In this paper, we propose an automatic bridge detection method in multispectral image to detect bridges over water. First, we extract the water region using NDWI index method by taking the advantage of the spectrum properties of water. Second, the water region is extended to include the bridges over water and the extended water region is adaptively segmented to locate the bridges roughly. Then, we separate the suspects into two kinds: Suspects along the bounder of the extended water region or the ones inside in this region, which are processed with different strategy. Partial interior suspects of the extended water region are extracted by the proposed RX detector. Finally, the context information of bridges is introduced to get the final decision. The experimental results on the TM images demonstrate the effectiveness of the proposed method.