OBJECT SEGMENTATION FOR LINEARLY POLARIMETRIC PASSIVE MILLIMETER WAVE IMAGES BASED ON PRINCIPLE COMPONENT ANALYSIS

Traditional passive millimeter wave imaging (PMMW) mechanism measures intensity-only radiometric energy of the scene, and the limited information restricts the subsequent process of target detection and recognition. Polarimetric phenomena provide an extra dimension of information and are utilized to improve the PMMW imaging performance. Based on linear polarization characteristics for terrain identification in our previous work, the horizontal, vertical and 45 degree linearly polarimetric images are obtained by manually changing the polarization orientation of the radiometer with a selfdesigned rotating installation. Then the related Stokes parameters and the linearly polarized angle are calculated for principal component analysis (PCA). Pixels with similar polarimetric characteristic are clustered in the score-plot feature space. Then the clusters are extracted to realize object segmentation of the raw image. Three types of objects including metallic stuff, lawn and concrete park are finally segmented, demonstrating that the proposed segmentation is feasible and effective.