Method for segmenting an image

1. Segmentation method of an image,

the method being computer-implemented and comprising the following steps:

(0) selecting pixels of said image;

(i) evaluating a curvature at each of said pixels,

a domain of values of the curvature being discretized into a finite number of values, and said curvature being evaluated in said discretized domain;

(ii) defining a topographic relief based on said image, an altitude of each of said pixels corresponding to said curvature evaluated in step (i);

(iii) calculating a watershed of said topographic relief defined in step (ii) to delimitate catchment basins;

(iv) defining a partition of said pixels based on said watershed calculated in step (iii);

(v) defining a segmentation of said image into one or more groups of pixels based on said partition defined in step (iv).

2. Segmentation method according to claim 1 , wherein step (0) consists in selecting all the pixels of said image.

3. Segmentation method according to claim 1 , wherein step (i) comprises:

a first sub-step of evaluating said curvature in a real bounded interval consisting of said domain; and

a second sub-step of discretizing said interval into a finite number of values.

4. Segmentation method according to claim 3 , wherein said interval corresponds to a closed bounded interval corresponding to [0,r] where r>0.

5. Segmentation method according to claim 1 , wherein said image is a 3D image.

6. Segmentation method according to claim 5 , wherein said 3D image is generated from a time-of-flight camera.

7. Method according to claim 6 , wherein step (i) comprises the following sub-steps:

(i.1) defining a collection of neighboring pixels for each pixel at which the curvature is to be evaluated;

(i.2) calculating a covariance matrix of said neighboring pixels;

(i.3) calculating eigenvalues of the covariance matrix calculated in step (i.2);

(i.4) evaluating said curvature based on the eigenvalues calculated in step (i.3).

8. Method according to claim 7 , wherein said curvature coincides with a value

λ

1

λ

1

+

λ

2

+

λ

3

,

λ 1 ≤λ 2 ≤λ 3 designating the eigenvalues calculated in step (i.3), ordered in increasing order.

9. Method according to claim 8 , wherein said collection of neighboring pixels defined in step (i.1) comprises the pixel at which the curvature is to be evaluated, and consists of at most eighty-one pixels substantially distributed in a square with a side of nine pixels, centred on the pixel at which the curvature is to be evaluated.

10. Method according to claim 9 , wherein said collection of neighboring pixels defined in step (i.1) consists of nine pixels substantially distributed in a square with a side of three pixels, centred at the pixel at which the curvature is to be evaluated.

11. Method according to claim 7 , wherein the step (i.3) also comprises a calculation of the eigenvectors of the covariance matrix calculated in step (i.2), and

wherein step (i) comprises the following sub-step:

(i.5) determining an orientation of the pixel at which the curvature is evaluated in step (i.4) based on the eigenvectors calculated in step (i.3).

12. Method according to claim 1 , wherein step (iii) comprises the following sub-steps:

(iii.1) determining a subset of said pixels for which the curvature evaluated in step (i) is locally minimal;

(iii.2) simulating a flooding of said topographic relief by a virtual rise of water from each of the pixels of said subset of said pixels;

(iii.3) calculating said watershed based on the simulation of step (iii.2).

13. Method according claim 1 , wherein step (iii) comprises the following sub-steps:

(iii.1′) simulating a flooding of said topographic relief by a virtual water runoff from each of said pixels;

(iii.2′) calculating said watershed based on the simulation of step (iii.1′).

14. Method according to claim 1 , further comprising a step of pre-processing said image prior to step (0), the pre-processing step comprising a filtering to reduce a noise of said image.

15. Method according to claim 1 , further comprising a post-processing step on groups of pixels of the segmentation defined in step (v).

16. A non-transitory computer-readable medium on which is stored a computer program comprising instructions which, when said computer program is executed, cause it to carry out the method according to claim 1 .