A novel landmine detection process using Karhunen Loeve Transform and Marker-based Watershed Segmentation in IR images

摘要

This paper presents a novel process for detecting landmines from Infrared (IR) images. The detection process uses Karhunen Loeve Transformation (KLT) to process a set of IR images collected at night. The detection process consist of number of stages. The first stage is the enhancement to improve the image quality. The next stage is the Marker-based Watershed Segmentation to specify the domain of possible objects. The final stage is the mine recognition. A comparison with Lee's method (Lee, 2001) is provided for three mine field with various soil types. Our method provided a higher detection rate and lower false alarms.