Objective: CT (Computed Tomography) is a useful tool to analyze object’s inner defect

its image more or less has noise. Crack is a common defect

with linear or piecewise linear characteristic. So

effectively detecting the linear crack in noisy image is the first important thing to analyze it. Method: A noisy CT linear crack detection algorithm based on fast Beamlet transform is presented in this paper. After analyzing the component and relation of Beamlet in mono-scale

a fast Beamlet transform is proposed. On the basis of it

a control variable about relativity is introduced. Then combining the tree-structure of Beamlet’s mutiscale and top to bottom inter-scale inhibition to optimize the object function

the linear crack is detected. Finally

considering the near pixels of detection result of linear crack

the edge of its domain is extracted. Result: In the numerical experiment

the images include original noisy CT image

adding Gaussian white noise of variance 0.1 to the original image

adding salt & pepper noise with density 0.1 to the original image. Compared with the method of Laplace

Canny or wavelet

the proposed method can detect the noisy CT linear crack effectively. Conclusion: Because Beamlet using lines to analyze the image data

the proposed method has robustness to noise. The detection result is convenient for next data analysis.