Fig. 1. (A) Double-wall radiographic technique with single-wall radiographic viewing. (B) Sampled radiograph from a pipe weld joint.

Fig. 2. The block diagram showing how the information flows through the components (blocks) of the object detection and recognition system [4].

Fig. 3. (A) Model image divided by n = 14 lines and columns, with the central point of reference Pcc. (B) The Pij neighborhood for a single (i, j) point.

Fig. 4. A matrix (3 × 3) Mref, derivatives δMref/δI and δMref/δj, and projections Ph (Mref) and Pv (Mref).

Fig. 5. (A) The test image—a typical weld joint radiograph. (B) The image k—a rectangular region extracted from the test image. (C) The model image.

Fig. 6. (A) The enclosing rectangle includes regions (A and B) with invalid reference points. (B) The enclosing rectangle does not include the whole weld length (L) to be inspected.

Fig. 7. (A) Bad sample (weld bead detection non-accepted by any term of the visual criterion). (B) Good sample (weld bead detection accepted by all terms of the visual criterion).

Fig. 8. Distribution of bad and good samples in a plot in terms of the matching parameters F and F_ph.

Fig. 9. The maximum, average, and minimum fitness value of population at each generation.

Fig. 10. (A) The length of the detected weld bead is incomplete. (B) The weld bead detection missed some part of the weld bead edge along the effective bead length (L).

Fig. 11. Image with 4 weld joint radiographs and results that were detected and recognized as weld beads.

Range of encoded variables in an individual

Results for different sets of matching parameters

Two test images and the results from genetic search and verification

Confusion table and the values for the performance measures

Results obtained in the test with multiples weld beads per image

Confusion table and the values for the performance measures including the both validation tests results