Research On Self-calibration Method Of Geometric Artifact For Nano-CT

X-ray computed tomography(CT)is an imaging technique that reconstructs the threedimensional structure of an object using the attenuation information of X-ray penetration at multiple angles.CT is widely used in many fields,such as biomedicine,non-destructive testing,and heritage conservation.With the increasing demand for imaging resolution for applications such as the analysis of new polymer materials and inspection of integrated circuits,X-ray CT imaging with the nanoscale resolution has become a research hotspot.Precise and stable system geometric parameters are essential for obtaining high-quality,highresolution 3D reconstructed images in nanocomputing systems.However,the small imaging field of view of nano-CT limits the application of body-model-based geometric parameter calibration methods.Studying the self-calibration technique of nano-CT geometric artifacts based on the object’s feature information is of great theoretical significance and application value.This paper focuses on the problem of self-calibration of geometric artifacts in nano-CT.The methodological research and application experiments are carried out from fast calibration of core parameters in the projection domain,iterative calibration of multiple parameters in the image domain and unsupervised blind calibration of geometric artifacts in the image domain,respectively.The main research contents and innovative results are summarized as follows:1.A self-calibration method of nano-CT geometric parameters based on mirror projection feature matching is proposed for the core parameter calibration problem in the projection domain.Small deviations of two core parameters,lateral offset,and rotation axis deflection angle can significantly affect the reconstructed image quality.The existing self-calibration method based on the gray projection value is vulnerable to the projection image noise and contrast,and the calibration accuracy can hardly meet the needs of the nano-CT system.To address the above problems,this paper proposes a self-calibration method based on feature point matching for nanoCT geometric parameters by using the matching features of paired projection images with a 180-degree difference as the mirror image of the rotation axis projection.First,the image alignment algorithm is used to coarsely extract and match the feature points of the mirrored projection image;then,using the feature of immobility of the rotation axis under the circular scanning trajectory,a false match point pair screening strategy based on feature triangles is designed to realize the fine screening of feature point pairs and improve the accuracy of feature matching;finally,the geometric parameters are calibrated by fitting the line where the rotation axis is located through RANSAC.Simulation and actual data results show that compared with the self-calibration method based on the gray projection value,the calibration results of the two core parameters of this paper are closer to the true value,the structural information of the detected object is better recovered,the reconstructed image is clearer,and the evaluation index value of image sharpness is improved by up to 25.2%.2.A self-calibration method of nano-CT geometric parameters based on the evaluation of geometric artifacts of the reconstructed images is proposed for the calibration of multiple parameters when the projection data is available.Geometric parameter mismatch leads to similar geometric artifacts,such as blurring and ghosting in the reconstructed images.Traditional geometric artifact characterization based on entropy,sharpness,and high-frequency energy has problems such as inaccurate characterization and poor generality,which are difficult to meet the demand for nano-CT geometric parameter calibration.To address the above problems,a selfcalibration algorithm of nano-CT geometric parameters based on an artifact evaluation network is proposed.First,based on the characteristics of geometric artifacts,a geometric artifact evaluation network based on the residual network model is designed to learn to characterize the geometric artifacts in the reconstructed images;then,an iterative calibration framework based on a genetic algorithm is designed to take the reconstructed images containing geometric errors as the input,and an adaptation function based on the output probability of the artifact evaluation network is constructed to guide the genetic algorithm to iteratively solve,so as to realize the system geometric The proposed algorithm is used as the input to construct an adaptation function based on the output probability of the artifact evaluation network and to guide the iterative solution of the genetic algorithm so as to achieve the geometric calibration of the system.Simulation and actual data results show that the proposed method can achieve accurate calibration of multiple geometric parameters of the nano-CT system and significantly suppress the blurring and ghosting of the reconstructed images caused by geometric artifacts.Compared with the existing methods,the image sharpness evaluation index is improved by up to 16.0%.3.A blind calibration method for geometric artifacts in the image domain based on unmatched data-driven nano-CT reconstructed images is proposed to address the problem of geometric artifact calibration in the image domain where projection data is unavailable.In the case of unavailable projection data,it is difficult to produce matched data sets.Furthermore,using a small number of single datasets is prone to overfitting problems when training network models with large parameter scales such as GAN.This paper proposes a blind calibration method based on non-matching datadriven geometric artifacts of nano-CT.The method uses Cycle GAN as the base network.It introduces residual blocks with Fourier transform according to the change of high and lowfrequency feature information of reconstructed images caused by geometric artifacts,which improves the characterization effect of geometric artifact features and achieves blind calibration of geometric artifacts under unmatched data sets.The simulation and actual data calibration results show that the method effectively solves the problem of blurring and ghosting of the reconstructed image caused by geometric artifacts.The original details of the reconstructed image are effectively recovered,realizing the blind calibration of geometric artifacts of the reconstructed image without projection data.Compared with the original Cycle GAN network,the image sharpness evaluation index is improved by up to 12%.