Study On Industrial CT3D Image Reconstruction And Segmentation Algorithm

Industrial Computed tomography (CT) uses x-ray to produce two-dimensional orthree-dimensional representation of the scanned object both externally and internally.Industrial CT has been used in many industrial areas for internal imspection ofcomponents. Its key uses have been flaw detection, failure analysis, metrology,assembly analysis. Cone-beam CT (CBCT), with high density flat plane detector, can beused for fast volumetric imaging with isotropic high-resolution and high use efficient ofx-ray source. Cone-beam CT mainly includes circular trajectory CBCT and helicaltrajectory CBCT. This dissertation mainly research improvement of industrialcone-beam CT reconstruction algorithm and3-D CT image defect segmentation.Due to the problems of technological production, the image collected by flat paneldetector has nonuniformity that cannot be eliminated by averaging multiplemeasurements, can be eliminated by correction. Nonuniformity exists in the whole flatpanel detector. The device is larger, the nonuniformity is more prominent. To solve thisproblem, this paper proposes non-uniform cone beam CT reconstruction algorithm forcorrection. Before the process of reconstruction, the nonuniformity correction isperformed on projection data. Using simulated projection and real projection to validatethe improved algorithm, the results show that the proposed algorithm can significantlyimprove the quality of the reconstructed image.Due to X-ray scattering and electronic noise, a three-dimensional projection dataobtained may contain large amounts of noise. If the projection data are directly used toreconstruct image, the SNR of image can be reduced. And Katsevich-type exactreconstruction algorithm of the helical cone-beam CT need to calculate the derivative ofthe projection data. The derivation of projection data with noise will enlarge the impactof noise in the projection data. The quality of the reconstructed image may be seriousreduced. Therefore the noisy projection data is necessarily filtered before derivation.Conventional median filter can effectively inhibit salt and pepper noise of the image,but it could damage some important details. This paper employs shock filter and itsimproved algorithm to denoise. Then the filtered projection data are used to reconstructimage through Katsevich algorithm. Experiment results proved that the imagereconstructed by our method is significantly better than the image reconstructed withmedian filter. With the development of cone-beam CT, the acquired data is larger and larger. Theimage reconstrucion become a very time-consuming task. GPU cluster is highperformance computer equipment that utilizes multiple computer resources to process acomputational task simultaneously. Thereby computation time is greatly reduced. In thisthesis, we use GPU cluster technology to speed up the reconstruction while preservingthe image quality. The parallelism of FDK reconstruction algorithms is investigated inthis work. A GPU cluster is constructed and used to conduct the experiment. A two-steptask distribution strategy was proposed. This parallel algorithm uses the samereconstruction formula as the sequential counterpart, which gives an identical imageresult. The results showed that the image quality was largely preserved while thecomputational time was greatly reduced. The image is larger, the speedup is greater.Defect segmentation of industrial CT volume data is an important task forapplication of industrial CT. In order to impliment automatic and fast segmentation ofthe defect, we proposed a defect segmentation of CT image approach combingmorphological methods with resample in polar coordinates technique. Firstly amorphological top-hat transformation is used to find the initial defects from the originalCT volume data. Secondly resample method is used for eliminating artifacts. Theexperimental results of segmentation proved that our method can accurately segment thedefects in three dimension CT images.