Research On The Key Problems Of Volume Rendering Based On Industrial CT Slices

As recognized as the best nondestructive testing technology in the world, industrial CTtechnology has now spread more widely in the industrial fields. The volume rendering ofindustrial CT slices, which can realize industrial CT three-dimensional testing, is animperative technology in the industry. It can avoid detection mistakes being brought in thetesting result by the subjective factors of the engineers. Because of the universality thatindustrial CT volume data usually contain noise and artifacts, several key techniques of directvolume rendering of ICT slices are studied in this dissertation, including denoising method forindustrial CT volume data based on NL means, optical transfer function design method involume rendering of industrial CT data facing to character of volume data and renderedresults, visualization algorithm based improved octree and CUDA and line feature extractionmethod for ICT data.Being geared to the key scientific problem that the denoising algorithm can effectivelyremove the interference brought by noise while preserve the detail information intactly. Someimproved effective methods based on NL means are researched in this paper. The computationof similarity in the conventional NL means algorithm is usually affected by noise. In order tosolve this problem, SSIM, Tchebichef moment and NSCT are applied for improving theaccuracy of similarity between sub-blocks. The experimental results show that the proposedmethods in the dissertation can not only have a good denoising performance but alsooutstanding detail-keeping ability.Efficient and accurate design methods of optical transfer function is another scientificproblem in volume rendering. Also it is a key technology. In this thesis design methods ofoptical transfer function are studied deeply from two perspectives, which are data drivenmethod and image driven methods. The satisfactory rendered results for industrial CT volumedata with slight artifacts and noise are provided by proposed algorithm based on Fisherstandards. However, histogram which usually lose spatial information are applied in theFisher algorithm. Thus, a kernel estimation based method of transfer function designing isproposed to solve this problem. Meanwhile, the optical transfer function design is changedinto function optimization problems, using the evaluation index such as entropy of the rendered image. The experimental results show that good visual effect can be obtained by thismethod.In a general way, if volume rendering of ICT data would be applied more widely in theindustrial field, there are two factors: one is a fast rendering time, another is a suitablehardware. So several most commonly used methods of volume rendering of the in theindustry are studied in this dissertation. The principles of four kinds of volume renderingalgorithm of industrial CT are analyzed and realized. Then performance of these methods iscompared adequately. Based on these, a fine grit parallel volume rendering algorithm basedon improved octree is put forward in Chapter4. And it is implemented with CUDA in theGPU. The experimental results show that, the rendering time of proposed method iscontracted obviously with little loss of rendering quality. It is more suitable for real-timerendering.At last, in order to obtain presentation of ICT data with sparse form linear featureextraction is studied in this paper. Linear feature of ICT is extracted by fast Wedgelettransform with information fusion of assistant sequence. The experimental results indicate theproposed algorithm can extract linear feature from volume data of ICT rapidly and effectively.Also, surface features are suppressed at the same time.