Fig. 1. Results of saliency detection. (a)-(d) (i)-(l) Partial images of original image; (e)-(h) (m)-(p) corresponding partial images of saliency map

Fig. 2. Segmentation results of reference hole contours. (a) Before contour segmentation; (b) after contour segmentation

Fig. 3. Diagrams of algorithm

Fig. 4. Pixel-level contour of reference hole

Fig. 5. Subpixel-level contour of reference hole

Fig. 6. Ransac fitting process and results. (a) Mask1; (b) Mask2; (c) Maskres; (d) result of ellipse fitting

Fig. 7. Simulated experiment site

Fig. 8. Detection results of reference hole. (a)-(g) Original images; (h)-(n) subpixel-level contours; (o)-(u) results of ellipse fitting

Fig. 9. Cropped composite images

Fig. 10. Cropped composite images added with 20% pepper noise

Fig. 11. Positioning error of composite images

Fig. 12. Experiment sample

Fig. 13. Measurement results of reference hole