New image reconstruction algorithm for CCERT: LBP + Gaussian mixture model (GMM) clustering

Abstract

Abstract This work focuses on the study of the image reconstruction algorithm of capacitively coupled electrical resistance tomography (CCERT). With the combination of a linear back projection (LBP) algorithm and an unsupervised Gaussian mixture model (GMM) algorithm, a new image reconstruction algorithm for CCERT is proposed. The LBP algorithm is used to implement the initial image reconstruction. The GMM algorithm is adopted to acquire the gray level threshold which will be used for the establishment of the gray level threshold filter. The final reconstructed image can be obtained with the thresholding operation. With a developed 12-electrode CCERT prototype system, the new image reconstruction algorithm is tested in image reconstruction experiments. The experimental results show that the proposed new image reconstruction algorithm is effective. The image reconstruction results are satisfactory. Compared with the conventional image reconstruction algorithms, the new image reconstruction algorithm (LBP + GMM) can obtain better reconstructed images with smaller relative image errors. It can obtain the reconstructed images with fewer empirical preset parameters and less manual intervention. In addition, with the introduction of the GMM algorithm, a relatively more suitable and reasonable gray level threshold can be obtained because the GMM algorithm implements the clustering process by utilizing both mean and variance information on the gray level distribution. Thus, better image reconstruction results can be obtained.