An image segmentation algorithm based on kernel estimation and moment-preserving

Industrial computed tomography (ICT) is a device which examines the internal structures of industrial components without destroying them. However, lots of artifact and noise usually exist in the industrial CT image and troubles are brought on the visualization and classification of industrial CT volume data at the same time. Image segmentation for Industrial CT images is one of the most important and fundamental tasks and techniques based on image threshold are typically simple and computationally efficient. However, the image segmentation results depend heavily on the threshold. The goal of our work is to decrease the noise and artifact in the industrial CT image by segmentation. Non-parameter estimation method is integrated to estimate the spatial probability distribution of gray-level values, and a criterion function by the moment-preserving is used. By optimizing the criterion function, an optimal global threshold is obtained. The experimental results for Industrial CT images demonstrate the success of the proposed image threshold method, as compared with the OTSU method and the moment-based method, which cannot segment the industrial CT images with low contrast ratio, much noise and artifact.