Method of pupil segmentation
View Patent ↗A method of pupil segmentation in a digital image of a vertebrate eye, said image being an intensity image composed of pixels having each a specific intensity value, the method comprising the steps of: deriving a texture image from the intensity image, said texture image being composed of pixels having each a specific contrast value; forming a combined image by point-wise combining the intensity image with the texture image, identifying a set of pixels in the combined image which fulfill a combined low-intensity and low-contrast criterion; and approximating a boundary of said set by a convex curve and taking said convex curve as a boundary of the pupil.
1. A method of pupil segmentation in a digital image of a vertebrate eye, said image being an intensity image composed of pixels having each a specific intensity value, the method comprising the steps of:
deriving a texture image from the intensity image, said texture image being composed of pixels having each a specific contrast value;
forming a combined image by point-wise combining the intensity image with the texture image,
identifying a set of pixels in the combined image which fulfill a combined low-intensity and low-contrast criterion; and
approximating a boundary of said set by a convex curve and taking said convex curve as a boundary of the pupil.
2. The method according to claim 1 , wherein said convex curve is a convex hull of said set of pixels.
3. The method according to claim 1 , wherein said combined image is a product image formed by point-wise multiplication of the intensity image with the texture image, and said combined low-intensity and low-contrast criterion is a threshold criterion for the product image.
4. The method according to claim 3 , wherein a threshold for said threshold criterion is determined individually for the digital image, based on a histogram analysis of the product image.
5. The method according to claim 4 , wherein the histogram analysis comprises identifying a peak at a lower end of a range of product values appearing in the product image, and the threshold is selected to separate said peak from the rest of the histogram.
6. The method according to claim 5 , wherein said histogram analysis comprises a filtering operation for smoothening the histogram, said filtering operation being performed before the peak is identified.
7. The method according to claim 1 , comprising a check whether a dimension of said boundary of the pupil, as described by said convex curve, is within a predetermined range.
8. The method according to claim 7 , comprising a step of measuring a distance between the eye and a camera with which the digital image is captured, and a step of scaling the intensity image in proportion to the distance between the camera and the eye, such that dimensions of a detail in the image of the eye indicate the corresponding true dimensions of the detail in the real eye.
9. The method according to claim 7 , wherein the segmentation process is aborted and the image is rejected when the dimension of the pupil boundary is not in said predetermined range.
10. The method according to claim 7 , wherein, when said dimension is not in the predetermined range, the steps of identifying a set of pixels fulfill the combined low-intensity and low-contrast criterion and of approximating the boundary of said set by a convex curve are repeated for a combined low-intensity and low-contrast criterion.
11. The method according to claim 10 , wherein the segmentation process is aborted and the image is rejected when the dimension of the pupil boundary is not in the predetermined range after the combined low-intensity and low-contrast criterion has been modified a predetermined number of times.
12. The method according to claim 1 , for a digital image that includes a specularity pattern caused by reflection of at least one light source at a cornea of the eye, the method comprising a step of removing said specularity pattern from the intensity image before the other steps of the segmentation are performed.
13. The method according to claim 12 , wherein an inpainting algorithm is used for removing the specularity pattern.
14. The method according claim 1 , wherein said convex curve is a closed geometrical line object that can only be described by more than five independent parameters.
15. The method according to claim 13 , wherein said line object is a polygon approximating a smooth closed curve.
16. A method of creating a binary iris code from a digital image of a vertebrate eye, including humans, with pupil segmentation, wherein said image is an intensity image composed of pixels having each a specific intensity value, the method comprising the steps of:
deriving a texture image from the intensity image, said texture image being composed of pixels having each a specific contrast value;
forming a combined image by point-wise combining the intensity image with the texture image;
identifying a set of pixels in the combined image which fulfill a combined low-intensity and low-contrast criterion;
approximating a boundary of said set by a convex curve and taking said convex curve as a boundary of the pupil; and
transforming a digital image of the iris into a coordinate system in which each point on the iris is described by a first coordinate that indicates the position of the point along the boundary of the pupil and a second coordinate indicates the distance of the point from said boundary.
17. An iris code generation kit comprising a camera and software and hardware suitable for carrying out the method of pupil segmentation in a digital image of a vertebrate eye, said image being an intensity image composed of pixels having each a specific intensity value, the method comprising the steps of:
deriving a texture image from the intensity image, said texture image being composed of pixels having each a specific contrast value;
forming a combined image by point-wise combining the intensity image with the texture image;
identifying a set of pixels in the combined image which fulfill a combined low-intensity and low-contrast criterion; and
approximating a boundary of said set by a convex curve and taking said convex curve as a boundary of the pupil.
18. An identification system for identifying vertebrates, including humans, on the basis of an iris code match, comprising at least one iris code generation kit comprising a camera and software and hardware suitable for carrying out the method of pupil segmentation in a digital image of a vertebrate eye, said image being an intensity image composed of pixels having each a specific intensity value, the method comprising the steps of:
deriving a texture image from the intensity image, said texture image being composed of pixels having each a specific contrast value;
forming a combined image by point-wise combining the intensity image with the texture image;
identifying a set of pixels in the combined image which fulfill a combined low-intensity and low-contrast criterion; and
approximating a boundary of said set by a convex curve and taking said convex curve as a boundary of the pupil.
19. A non-transitory computer readable medium storing a computer program to perform the method of claim 1 .