Terahertz image segmentation of occluded objects based on mean clustering

Terahertz imaging of occluded objects is a popular technology, which is an important feature superior to visible light imaging. However, the quality of the target image is obviously damaged due to factors such as discontinuity and absorption of the occlusion, which makes it difficult to segment the image. It is especially difficult to segment digital holographic reconstruction images of small objects with high resolution. In this paper, clustering segmentation algorithms are compared for terahertz images of metal objects which is occluded by paper. K-means, fuzzy C-means (FCM) and fuzzy c-means clustering with spatial constraints (FCM-S) algorithms were used respectively. Since the minimum horizontal target is only three pixels, the mean template size in these algorithms is all 3∗3. The experimental results show that FCM-S of the segmentation effect obtained is the best among the three algorithms, because FCM-S considers the pixel neighborhood information.