Robust dislocation defects region segmentation for polysilicon wafer image

The segmentation of dislocation defects has important value for photovoltaic polysilicon wafers to predict power generation efficiency and optimize the growth process of silicon ingots. However, the accurate segmentation of dislocation defects has still been a challenging issue because the intensity and shape feature between dislocation and background in the polysilicon wafer image show significant similarities. Thus, a new globalization framework based on Atmospheric Scattering Model (ASM) is proposed to accurately complete the dislocation defects region segmentation. Firstly, a parameter-optimized ASM was proposed to enhance the contrast of PL images and highlight the dislocation features in the PL images. Secondly, the contour detector combines multi-scale probability of boundaries (mPb) into the globalization framework based on spectral probability of boundaries (sPb) in this paper. Finally, an oriented watershed transform combined (owt) with an ultrametric contour map (ucm) method effectively removes the artifacts of potential contours and achieves an accurate segmentation of dislocation defects in the images.