Selecting features from image data

• US 9,576,218 B2
• Filed: 11/04/2014
• Issued: 02/21/2017
• Est. Priority Date: 11/04/2014
• Status: Active Grant

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Huang, Hung Khei, Denney, Bradley Scott
• Primary Examiner(s): Park, Chan
• Assistant Examiner(s): Thomas, Mia M
• Application Number:
  US14/532,625
• Publication Number:
  US 20160125260A1
• Time in Patent Office: 840 Days
• US Class Current: 1/1
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

SIFT features are selected from an input image. A SIFT procedure is applied to the input image to obtain candidate keypoints. For each candidate keypoint, there is calculation of a first Laplacian value (L_u) for pixels in an upper Scale Space and a second Laplacian value (L_l) for pixels in a lower Scale Space, based on the position of the candidate keypoint. A candidate keypoint is discarded if a Laplacian value L_c of the keypoint position is less than or equal to either of L_u or L_l. In the case that the candidate keypoint is not discarded, the candidate keypoint'"'"'s Laplacian strength (L_s) is calculated, based on a relative change in Laplacian value from L_c to L_u and from L_c to L_l. One or more candidate keypoints are selected as SIFT features based on the corresponding Laplacian strength.

Claims

1. A method for selecting SIFT features from an input image, the method comprising:

applying a SIFT procedure to the input image to obtain candidate keypoints;

calculating, for each candidate keypoint, a first Laplacian value (Lu) for pixels in an upper Scale Space and a second Laplacian value (Ll) for pixels in a lower Scale Space, based on the position of the candidate keypoint;

discarding a candidate keypoint if a Laplacian value Lc of the keypoint position is less than or equal to either of Lu or Ll;

calculating, in the case that the candidate keypoint is not discarded, the candidate keypoint'"'"'s Laplacian strength (Ls) based on a relative change in Laplacian value from Lc to Lu and from Lc to Ll; and

selecting one or more candidate keypoints as SIFT features based on the corresponding Laplacian strength.

2. The method according to claim 1, wherein Ls is calculated at least in part from Lu and one or more of Lc/Lu and Lc/Ll.

3. The method according to claim 2, wherein Ls is calculated by Ls=(((Lc−

Ll)/Ll)+((Lc−

Lu)/Lu))/2.

4. The method according to claim 2, wherein Ls is calculated by Ls=[Lc−

max(Ll, Lu)]/max(Ll, Lu).

5. The method according to claim 2, wherein Ls is calculated by Ls=[Lc −

min(Ll, Lu)]/min(Ll, Lu).

6. The method according to claim 2, wherein Ls is calculated by Ls=Lc/max(Ll, Lu).

7. The method according to claim 1, wherein the candidate keypoint is added to a list of keypoints sorted by descending Laplacian strength.

8. The method according to claim 7, wherein the selection comprises selecting a number of top ranked keypoints from the list.

9. The method according to claim 8, wherein the number of keypoints selected from the list is a specified percentage of the total number of keypoints.

10. The method according to claim 8, wherein the number of keypoints selected from the list is a predetermined fixed maximum number of keypoints.

11. The method according to claim 10, wherein the predetermined fixed maximum number of keypoints is based on user input.

12. The method according to claim 1, wherein the selected keypoints are used to search for a corresponding image.

13. An apparatus for selecting SIFT features from an input image, comprising:

a computer-readable memory constructed to store computer-executable process steps; and

a processor constructed to execute the process steps stored in the memory,

wherein the process steps cause the processor to;

apply a SIFT procedure to the input image to obtain candidate keypoints;

calculate, for each candidate keypoint, a first Laplacian value (Lu) for pixels in an upper Scale Space and a second Laplacian value (Ll) for pixels in a lower Scale Space, based on the position of the candidate keypoint;

discard a candidate keypoint if a Laplacian value Lc of the keypoint position is less than or equal to either of Lu or Ll;

calculate, in the case that the candidate keypoint is not discarded, the candidate keypoint'"'"'s Laplacian strength (Ls) based on a relative change in Laplacian value from Lc to Lu and from Lc to Ll; and

select one or more candidate keypoints as SIFT features based on the corresponding Laplacian strength.

14. The apparatus according to claim 13, wherein Ls is calculated at least in part from Lu and one or more of Lc/Lu and Lc/ Ll.

15. The apparatus according to claim 14, wherein Ls is calculated by Ls=(((Lc−

Ll)/Ll)+((Lc−

Lu)/Lu))/2.

16. The apparatus according to claim 14, wherein Ls is calculated by Ls=[Lc−

max(Ll, Lu)]/max(Ll, Lu).

17. The apparatus according to claim 14, wherein Ls is calculated by Ls=[Lc−

min(Ll, Lu)]/min(Ll, Lu).

18. The apparatus according to claim 14, wherein Ls is calculated by Ls=Lc/max(Ll, Lu).

19. The apparatus according to claim 13, wherein the candidate keypoint is added to a list of keypoints sorted by descending Laplacian strength.

20. A non-transitory computer-readable storage medium storing computer-executable process steps for causing a machine to perform a method for selecting SIFT features from an input image, the method comprising:

applying a SIFT procedure to the input image to obtain candidate keypoints;

calculating, for each candidate keypoint, a first Laplacian value (Lu) for pixels in an upper Scale Space and a second Laplacian value (Ll) for pixels in a lower Scale Space, based on the position of the candidate keypoint;

discarding a candidate keypoint if a Laplacian value Lc of the keypoint position is less than or equal to either of Lu or Ll;

calculating, in the case that the candidate keypoint is not discarded, the candidate keypoint'"'"'s Laplacian strength (Ls) based on a relative change in Laplacian value from Lc to Lu and from Lc to Ll; and

selecting one or more candidate keypoints as SIFT features based on the corresponding Laplacian strength.