SYSTEM AND METHOD FOR AUTOMATIC DETECTION OF ANATOMICAL FEATURES ON 3D EAR IMPRESSIONS

• US 20100103170A1
• Filed: 10/22/2009
• Published: 04/29/2010
• Est. Priority Date: 10/27/2008
• Status: Active Grant

• Current Assignee: Siemens Audiologische Technik GmbH (Siemens AG)
• Original Assignee: Siemens Corp. (Siemens AG)
• Inventors: Zouhar, Alexander, Baloch, Sajjad Hussain, Fang, Tong, Azernikov, Sergei, Melkisetoglu, Rupen
• Granted Patent: US 8,229,180 B2
• Time in Patent Office: Days
• US Class Current: 345/420
• CPC Class Codes: G06V 20/64 Three-dimensional objects
• CPC Class Codes: H04R 2225/77 Design aspects, e.g. CAD, of hearing aid tips, moulds or housings
• CPC Class Codes: H04R 25/652 Ear tips; Ear moulds hybrid ear moulds or post-processing thereof for their customisation H04R25/659
• CPC Class Codes: H04R 25/658 Manufacture of housing parts

Abstract

A method for detecting anatomical features in 3D ear impressions includes receiving a 3D digital image of a 3D ear impression, obtaining a surface of the ear impression from the 3D image, analyzing the surface with one or more feature detectors, the detectors adapted to detecting generic features, including peak features, concavity features, elbow features, ridge features, and bump features, and derived features that depend on generic features or other derived features, and forming a canonical ear signature from results of the detectors, where the canonical ear signature characterizes the 3D ear impression.

Claims

1. A computer implemented method for detecting anatomical features in 3D ear impressions, the method performed by said computer comprising the steps of:

receiving a 3D digital image of a 3D ear impression;

obtaining a surface of said ear impression from said 3D image;

analyzing said surface with one or more feature detectors, said detectors adapted to detecting generic features, including peak features, concavity features, elbow features, ridge features, and bump features, and derived features that depend on generic features or other derived features; and

forming a canonical ear signature from results of said detectors, wherein said canonical ear signature characterizes said 3D ear impression.

2. The method of claim 1, wherein detecting a peak feature comprises:

orienting said surface in a local coordinate system;

initializing a scanning plane whose normal is parallel to a z-axis of said coordinate system;

forming a first level set formed by intersecting said surface with said scanning plane;

forming a second level set by intersecting said surface with a scanning plane displaced in said normal direction along said surface; and

analyzing said first and second level sets to search for critical points enclosed by said level set, wherein the existence of a critical point indicates the presence of a peak feature, wherein peak features of an ear include the canal tip, concha peak, and helix peak.

3. The method of claim 2, further comprising, when no critical point was detected, displacing the scanning plane in the normal direction for a new second level set and using the previous second level set as a new first level set, and repeating said step of analyzing said first and second level sets to search for critical points.

4. The method of claim 2, wherein analyzing said first and second level sets comprises:

comparing a number of connected components in each level set, wherein each connected component is a contour;

identifying the level set enclosing a greater number of contours as a lower level set;

identifying and associating corresponding contours included in the two level sets; and

comparing an arc-length of any unassociated contour in the lower level set to a tolerance, wherein a contour whose arc-length is less than the tolerance encloses a critical point.

5. The method of claim 1, wherein detecting a concavity feature comprises:

identifying a region of interest (ROI) contained in said surface;

specifying a first and second orthogonal directions along said ROI;

scanning said ROI along said first direction to generate a profile of first contours in said ROI;

identifying those points on each first contour having a lowest negative curvature;

computing a seed point location from an average location of said points having a lowest negative curvature;

scanning said ROI along said second direction to generate a profile of second contours;

identifying those points on each second contour having a lowest negative curvature;

correcting the seed point location by averaging over the locations of said points having a lowest negative curvature on each second contour; and

growing a concave regions outward from said seed point,

wherein concavity features of an ear include the tragus center, the anti-tragus center, the crus center, and the anti-helix center.

6. The method of claim 5, further comprising repeating steps of scanning said ROI along said first direction, identifying those points on each first contour having a lowest negative curvature, averaging said seed point location over locations of said points, scanning said ROI along said second direction, identifying those points on each second contour having a lowest negative curvature, and averaging the seed point location over the locations of said points, until said seed point location converges.

7. The method of claim 1, wherein detecting an elbow feature comprises:

estimating a centerline of said shell surface;

constructing a plurality of cross sectional profile contours oriented along said centerline;

forming radial contours connecting corresponding points across said set of profile contours;

analyzing each radial contour to identify points having a high curvature;

discarding those points whose curvature falls below a predefined threshold; and

analyzing all point combinations of the remaining point that define a cross sectional plane of said shell, and clustering said points in a generalized Hough space to find a plane that yields a minimal variation as identifying an elbow in said shell,

wherein elbow features of an ear include the first bend and the second bend.

8. The method of claim 1, wherein detecting a ridge feature comprises:

finding a starting point and an ending point of a ridge on said surface; and

computing a geodesic connecting said starting and ending points that minimizes a cost functional of a curve connecting said starting and ending points wherein the said cost functional is constructed as the length of the geodesic weighted by a decreasing function of curvature of each point along said curve, wherein said ridge is defined by said geodesic,

wherein ridge features of an ear include the intertragal notch, the crus-side canal ridge, and the crus-helix ridge.

9. The method of claim 1, wherein detecting a bump feature comprises:

constructing a plurality of cross sectional profile contours along said surface;

selecting a contour at one end of said plurality of contours as a reference contour;

for each other contour in said plurality of contours,

comparing an area of the reference contour with the area of a current contour in said plurality of contours;

finding bump points on said current contour, wherein an area enclosed by said current contour exceeds that of said reference contour by a predetermined threshold;

and

collecting said bump points into areas by region growing from each bump point, wherein bump features of an ear include the canal bulbous.

10. The method of claim 9, further comprising updating said reference contour with a current contour, wherein a difference of areas enclosed by said current contour and of said reference contour is less than said predetermined threshold.

11. The method of claim 9, wherein finding bump points on said current contour comprises:

projecting points on the current contour onto the plane of the reference contour;

finding a closest point on the reference contour for each projected current point, to form a pair of corresponding points;

comparing a distance of a projected current point to a projected centroid of the current contour to a distance of the corresponding reference point to a centroid of the reference contour, for each corresponding point pair; and

determining that the projected current point is bump if the distance to its centroid is greater than the corresponding distance of the reference contour point by a predetermined threshold.

12. The method of claim 9, wherein an area enclosed by a contour is calculated from

13. The method of claim 9, wherein region growing for each bump point comprises collecting convex surface points with curvature greater than a predefined threshold.

14. A computer implemented method for detecting anatomical features in 3D ear impressions, the method performed by said computer comprising the steps of:

receiving a 3D digital image of a 3D ear impression;

obtaining a surface of said ear impression from said 3D image;

scanning a canal region of said surface with horizontal planes to yield a plurality of cross sectional contours;

identifying and rejecting outlying contours near a tip of said canal, by comparing the circumference and the position of extremal points on a first principle axis of said outlying contour with the circumference and corresponding extremal point positions on a first principle axis of an inner contour;

analyzing the remaining contours for ellipticity, and further analyzing highly elliptical contours to identify potential starting ridge points;

finding a pair of starting points from said potential ridge points;

finding an ending point at the base of the shell at the center of convexity of a contour connecting the tragus center point and the anti-tragus center point;

detecting a ridge between each pair of starting points and said ending point;

classifying a shorter ridge as an inter-tragal notch, and identifying the other starting point as the starting point of a crus-side canal ridge; and

detecting a ridge as the shortest of the geodesics from the starting point of a crus-side canal ridge to the each point on the bottom boundary of the shell and identifying this ridge as a crus-side canal ridge.

15. The method of claim 14, further comprising:

detecting a concha center point, a tragus center point, and a ridge in said surface;

calculating a weighted geodesic connected said concha center point and said tragus center point, wherein said geodesic is weighted by a curvature function that is a decreasing function of curvature and that penalizes an optimal geodesic from passing through points on said ridge; and

finding an intersection point of said geodesic with said ridge, wherein said ridge is one of crus-side canal ridge and the intertragal notch, and said intersection point is one of a canal-concha intersection point and a canal-crus intersection point.

16. The method of claim 14, further comprising:

detecting a helix peak in said surface;

finding a contour on said surface that defines a bottom of an opening;

initializing an optimal arc-length; and

for each point on said contour,

finding a weighted geodesic on said surface from said point to said helix peak, wherein said geodesic is weighted by a curvature function that is a decreasing function of curvature; and

calculating an arc-length of said geodesic, wherein if said geodesic arc-length is less than said optimal arc-length, assigning said geodesic arc-length to said optimal arc-length,

wherein a geodesic with a shortest optimal arc-length is identified as the crus helix ridge of said ear.

17. The method of claim 14, further comprising:

detecting the helix peak point, the concha center point, the crus helix ridge, and the crus-side canal ridge in said surface;

finding an unweighted helix-concha geodesic on said surface connecting said helix peak point and said concha center point;

finding a crus pivot point as the lowest point on said helix-concha geodesic;

finding a weighted pivot-helix bottom geodesic on said surface connecting a bottom point of said crus helix ridge and said crus pivot point, wherein said weights favor concave points;

finding a weighted pivot-canal bottom geodesic on said surface connecting said crus pivot point and a bottom point of said crus-side canal ridge, wherein said weights favor concave points;

selecting a seed point in an area enclosed by said pivot-canal bottom geodesic and said pivot-helix bottom geodesic; and

growing a region from said seed point bounded by said said pivot-canal bottom geodesic and said pivot-helix bottom geodesic, wherein said area is identified as the crus area.

18. The method of claim 14, further comprising:

detecting the crus center point, the concha center point, the tragus center point, the anti-tragus center point, and the crus valley plane in said ear impression image;

finding a crus-valley contour by intersecting said crus-valley plane with said surface, wherein said crus-valley contour is oriented wherein the beginning point of said contour is close to said tragus center point and the end point of said contour is close to said anti-tragus center point;

finding a seed point on said contour that minimizes a distance from said contour to said crus center point;

finding a segment of said contour that extends from said seed point to said contour end point;

determining a reference direction from a vector extending from the beginning point of said contour to the end point of said contour;

finding a first point as a point on said contour segment whose tangential inner product with said reference direction is below a predetermined threshold;

finding a shortest weighted geodesic from said concha center point to said contour, wherein said weights favor convex points;

finding a second point from the intersection of said geodesic with said contour segment; and

calculating crus-concha intersection from an average of said first point and said second point.

19. A program storage device readable by a computer, tangibly embodying a program of instructions executable by the computer to perform the method steps for detecting anatomical features in 3D ear impressions, the method comprising the steps of:

receiving a 3D digital image of a 3D ear impression;

obtaining a surface of said ear impression from said 3D image;

analyzing said surface with one or more feature detectors, said detectors adapted to detecting generic features, including peak features, concavity features, elbow features, ridge features, and bump features, and derived features that depend on generic features or other derived features; and

forming a canonical ear signature from results of said detectors, wherein said canonical ear signature characterizes said 3D ear impression.

20. The computer readable program storage device of claim 19, wherein detecting a peak feature comprises:

orienting said surface in a local coordinate system;

initializing a scanning plane whose normal is parallel to a z-axis of said coordinate system;

forming a first level set formed by intersecting said surface with said scanning plane;

forming a second level set by intersecting said surface with a scanning plane displaced in said normal direction along said surface; and

analyzing said first and second level sets to search for critical points enclosed by said level set, wherein the existence of a critical point indicates the presence of a peak feature, wherein peak features of an ear include the canal tip, concha peak, and helix peak.

21. The computer readable program storage device of claim 20, the method further comprising, when no critical point was detected, displacing the scanning plane in the normal direction for a new second level set and using the previous second level set as a new first level set, and repeating said step of analyzing said first and second level sets to search for critical points.

22. The computer readable program storage device of claim 20, wherein analyzing said first and second level sets comprises:

comparing a number of connected components in each level set, wherein each connected component is a contour;

identifying the level set enclosing a greater number of contours as a lower level set;

identifying and associating corresponding contours included in the two level sets; and

comparing an arc-length of any unassociated contour in the lower level set to a tolerance, wherein a contour whose arc-length is less than the tolerance encloses a critical point.

23. The computer readable program storage device of claim 19, wherein detecting a concavity feature comprises:

identifying a region of interest (ROI) contained in said surface;

specifying a first and second orthogonal directions along said ROI;

scanning said ROI along said first direction to generate a profile of first contours in said ROI;

identifying those points on each first contour having a lowest negative curvature;

computing a seed point location from an average location of said points having a lowest negative curvature;

scanning said ROI along said second direction to generate a profile of second contours;

identifying those points on each second contour having a lowest negative curvature;

correcting the seed point location by averaging over the locations of said points having a lowest negative curvature on each second contour; and

growing a concave regions outward from said seed point,

wherein concavity features of an ear include the tragus center, the anti-tragus center, the crus center, and the anti-helix center.

24. The computer readable program storage device of claim 23, the method to further comprising repeating steps of scanning said ROI along said first direction, identifying those points on each first contour having a lowest negative curvature, averaging said seed point location over locations of said points, scanning said ROI along said second direction, identifying those points on each second contour having a lowest negative curvature, and averaging the seed point location over the locations of said points, until said seed point location converges.

25. The computer readable program storage device of claim 19, wherein detecting an elbow feature comprises:

estimating a centerline of said shell surface;

constructing a plurality of cross sectional profile contours oriented along said centerline;

forming radial contours connecting corresponding points across said set of profile contours;

analyzing each radial contour to identify points having a high curvature;

discarding those points whose curvature falls below a predefined threshold; and

analyzing all point combinations of the remaining point that define a cross sectional plane of said shell, and clustering said points in a generalized Hough space to find a plane that yields a minimal variation as identifying an elbow in said shell,

wherein elbow features of an ear include the first bend and the second bend.

26. The computer readable program storage device of claim 19, wherein detecting a ridge feature comprises:

finding a starting point and an ending point of a ridge on said surface; and

computing a geodesic connecting said starting and ending points that minimizes a cost functional of a curve connecting said starting and ending points wherein the said cost functional is constructed as the length of the geodesic weighted by a decreasing function of curvature of each point along said curve, wherein said ridge is defined by said geodesic,

wherein ridge features of an ear include the intertragal notch, the crus-side canal ridge, and the crus-helix ridge.

27. The computer readable program storage device of claim 19, wherein detecting a bump feature comprises:

constructing a plurality of cross sectional profile contours along said surface;

selecting a contour at one end of said plurality of contours as a reference contour;

for each other contour in said plurality of contours,

comparing an area of the reference contour with the area of a current contour in said plurality of contours;

finding bump points on said current contour, wherein an area enclosed by said current contour exceeds that of said reference contour by a predetermined threshold;

and

collecting said bump points into areas by region growing from each bump point, wherein bump features of an ear include the canal bulbous.

28. The computer readable program storage device of claim 27, the method further comprising updating said reference contour with a current contour, wherein a difference of areas enclosed by said current contour and of said reference contour is less than said predetermined threshold.

29. The computer readable program storage device of claim 27, wherein finding bump points on said current contour comprises:

projecting points on the current contour onto the plane of the reference contour;

finding a closest point on the reference contour for each projected current point, to form a pair of corresponding points;

comparing a distance of a projected current point to a projected centroid of the current contour to a distance of the corresponding reference point to a centroid of the reference contour, for each corresponding point pair; and

determining that the projected current point is bump if the distance to its centroid is greater than the corresponding distance of the reference contour point by a predetermined threshold.

30. The computer readable program storage device of claim 27, wherein an area enclosed by a contour is calculated from

31. The computer readable program storage device of claim 27, wherein region growing for each bump point comprises collecting convex surface points with curvature greater than a predefined threshold.