Terahertz image segmentation using k-means clustering based on weighted feature learning and random pixel sampling

• A novel approach of segmentation of Terahertz images is proposed.
• We propose a weighted feature space and a simple random sampling in k-means clustering.
• The goal of the approach is to estimate the expected centers, select the relevant features and their scores, and classify the observed pixels.
• We propose to estimate automatically the random sample size and the selected feature number.
• Our approach is more appropriate for achieving the best compactness inside clusters, the best discrimination of features, and the best tradeoff between the clustering accuracy and the low computational cost.

Terahertz (THz) imaging is an innovative technology of imaging which can supply a large amount of data unavailable through other sensors. However, the higher dimension of THz images can be a hurdle to their display, their analysis and their interpretation. In this study, we propose a weighted feature space and a simple random sampling in k-means clustering for THz image segmentation. Our approach consists in estimating the expected centers, selecting the relevant features and their scores, and classifying the observed pixels of THz images. Automatic estimation of the random sample size and the selected feature number are also proposed in this paper. Our approach is more appropriate for achieving the best compactness inside clusters, the best discrimination of features, and the best tradeoff between the clustering accuracy and the low computational cost. Our approach of segmentation is evaluated by measuring performances and appraised by a comparison with some related works.