Automatic Parameter Estimation for the Discrete Algebraic Reconstruction Technique (DART)

Geometric and Visual Reconstruction of Binary Shapes

“…Van Aarle et al introduced an elegant solution for the automatic estimation of the intensity levels [104]. The main advantage of their method is that it is not restricted to binary images, and its efficiency for reconstructing images containing three unknown gray-levels was also demonstrated.…”

“…To overcome this issue, a Gaussian filter is applied to the intermediate solutions. While in the original description of the DART method this smoothing operation was performed only on the non-fixed pixels, in the more recent work of the authors [104] it was proposed to apply it on the whole image. We also found in our experiments that this latter approach provides more accurate results.…”

“…The main drawback of this approach is that it requires full DART reconstruction in each iteration step, therefore it is computationally inefficient. In [104], the DART method has been extended with a more advanced segmentation technique called Projection Distance Minimization (PDM) [11]. While in the original DART method the segmentation parameters are fixed, the PDM-DART method re-estimates these parameters in each iteration step by minimizing the distance between the projections of the intermediate segmented image and the projection data.…”

“…The strength of the noise can be characterized by the incident beam intensity (I 0 ), which is the maximal number of particle count in the beams. This noise model was recently used in [104], Here, we tested the methods for I 0 = 10 5 and I 0 = 10 4 , and the noiseless cases are referred by I 0 = ∞. As Fig.…”

“…To evaluate the reconstruction results we used the error measure called relative Number of Misclassified Pixels (rNMP), which gives the ratio of the number of erroneous pixels and the number of object pixels in the original phantom image. This error measure was recently used in [104] to evaluate the results of the PDM-DART method. The rNMP measures and runtime performances provided by the proposed method can be found in …”