A hybrid method based region of interest segmentation for continuous wave terahertz imaging

Abstract

Terahertz (THz) imaging is an innovative technology used in many fields such as security, biology and hidden target detection. Region of interest (ROI) segmentation is essential for applications of THz imaging. However, THz images are often severely degraded by motion blur, poor resolution and severe noise. In this paper, a hybrid ROI segmentation method is proposed for continuous-wave (CW) THz images. The method combines block matching 3D denoising, fuzzy c-means clustering, morphology operation and canny edge detection. The hybrid method is applied to a THz image of patterned plates to segment letters. The letters are correctly segmented with accuracy, sensitivity and specificity of 94.3%, 91.6% and 95.5% respectively. Moreover, a THz image of an ex vivo rat brain is taken. The tumor area is precisely segmented using this hybrid ROI segmentation method with accuracy, sensitivity and specificity of 95.6%, 84.5% and 97.7% respectively. It is suggested that the proposed hybrid ROI segmentation method can perform well for CW THz images and even THz biological images.