Research On Evaluation Of ET Reconstructed Image Quality Based On Cluster

Electrical Tomography (ET) is a new testing technology developed in 1980. Dueto the advantage of simple structure, low cost and non-radiant, ET technique is widelyused in the petrochemical and medical fields. ET reconstruction image qualityevaluation is one of the most important research directions of ET field, because it ishelpful to improve image reconstruction algorithm and hardware circuit. At presentthere are a lot of image reconstruction algorithms, such as sensitivity analysis, linearback projection, conjugate gradient method, etc. Their principles are different andthey apply in different scopes. As a result, a quantitative assessment index is neededto evaluate quality of reconstructed images. Many researchers have proposed someindices to evaluate ET reconstructed image quality, but they are limited. We havepresented clustering-based image reconstructed algorithm in previous research andobtained some achievements. This paper analyzes the mechanism of clustering-basedtomography, and cluster validity indices are used to evaluate the quality ofreconstructed images. The main work and results are as follows:1) This paper proposes a new cluster algorithm based on previous cluster methodsto reconstruct ET images, by extracting mean value、variance and second-orderdifference of origin vector to form vectors of three-dimension feature. By experimentit can be found that new cluster method has better reconstructed image quality.Holistic gray variance of image is presented to evaluate quality of reconstructedimage. The greater variance is, the better the quality of reconstructed image quality.This method is suitable for all kinds of image reconstructed algorithms. The feasibilityof the method is testified by experiments. This work lays a foundation for ETreconstructed image quality evaluation.2) This paper presents a new fuzzy gap statistics and makes two improvementsbased on previous fuzzy gap statistics: 1) redefine number of samples to each cluster;2) calculate class-center by overall membership degree. Modified gap statistics canwork out accurately center of each cluster and obtain optimal class number preciselyin some condition. 3) ET reconstructed image quality evaluation software platform is designed basedon GUI of MATLAB and it can provide convenience for further research of ETreconstructed image quality evaluation. Common fuzzy validity indices are placed thesame interface, which is friendly and easy to compare.