Fig. 1. An artist’s concept of the FORTÉ satellite. Note the long antenna (75 ft) as compared to the satellite body (7 ft).

Fig. 2. The average image of all the images used in the data set.

Fig. 3. Two carefully selected images revealing most of the artefacts regions.

Fig. 4. Mask constructed making a union of the result from thresholding methods applied on images presented inFig. 2 and Fig. 3.

Fig. 5. The 1D iterative extrapolation method approximating for a glitch in a sine function.

Fig. 6. A cut of the pixel intensity of an image used in the data set. The iterative extrapolation formula is then applied in one dimension in order to illustrate a sequence of two iterative steps. Note the randomly distributed pixel elements that are not classified as the artefact region and the original image information that is kept as a ‘ground truth’.

Fig. 7. Two examples showing the use of the extrapolation formula described in Section 2.2. To the left are shown the raw images with the antenna present. The figures to the right are the results of applying the iterative formula as given by (1), (2), (3a), (3b), (4a), (4b) and (4c).

Fig. 8. Left, example of an input image presented to the PCNN algorithm. Right, the corresponding accumulated response matrix. The black areas represent neurons firing only once, the lighter are neurons firing more than once.

Fig. 9. Binary outputs for the two PCNN iterations needed to perform edge detection. The stimuli S (input) is identical to the first PCNN output (left image).

Fig. 10. Left, the satellite image used as input. Right, the same image with edge detection result overlaid.

Table 1. The PCNN parameter values used when performing accumulated response (PCNN I), edge detection (PCNN II) and the number of iterations Na