High-Precision Image Reconstruction for Binary-Valued X-Ray CT From Limited-View Projections

In many industrial scenarios, the object to be imaged is binary valued and its composition can be known in advance. Typical X-ray computed tomography (CT) methods are challenged to achieve high quality images from limited projections. An image reconstruction method for high precision X-ray CT of binary-valued distributions is proposed for projections of limited views. A binary matrix of only two elements, i.e., 0 and 1, is introduced for target distributions with two known linear attenuation coefficients. Projection histograms along ray paths from different projection views are used to derive the binary matrix. The Sigmoid function is utilized to achieve the binary matrix in which each row has and only has one non-zero item of one. The distribution of the imaged object is obtained by mapping the linear attenuation coefficient at each grid one by one. Compared with SART and FBP methods, the proposed method yields lower normalized mean distances and higher image similarity from limited views. In noisy cases, the normalized mean absolute distance of the proposed method was reduced by more than 40%. In experiments, an aluminum column and carbon rod were imaged with clear edges and fewer artifacts by the proposed method. Also, it is used to reconstruct triple-valued distributions, with image errors lower than SART and FBP.