A Multi-Class Classification System for Passive Millimeter-Wave Image

Since passive millimeter wave (PMMW) imaging has a lot of advantages over optical and infa-red imaging, it has been widely applied in the areas regarding remote sensing, security check and military. In this paper, a content based passive millimeter wave image classification system is proposed. The high-dimensional adaptive manifolds and Laplacian of Gaussian based PMMW image filtering is firstly exploited to remove the strong noise existing in the PMMW images. The local binary pattern is applied to extract image feature for each training sample. The extracted features are then used to achieve histogram intersection matrix and fed to histogram intersection kernel support vector machine (IKSVM) to build the classifier. To validate the performance of the proposed system, a PMMW image database that contains 560 simulated PMMW images in three classes is also introduced.