Method Of Detecting Red-Eye Objects In Digital Images Using Color, Structural, And Geometric Characteristics

• US 20100172575A1
• Filed: 01/07/2009
• Published: 07/08/2010
• Est. Priority Date: 01/07/2009
• Status: Active Grant

• Current Assignee: Advanced Interconnect Systems Limited (Vector Capital Corporation)
• Original Assignee: Seiko Epson Corporation (Seiko Group)
• Inventors: Olekas, Christopher V., Ljubuncic, Bojan, Lukac, Rastislav
• Granted Patent: US 8,295,593 B2
• Time in Patent Office: Days
• US Class Current: 382/164
• CPC Class Codes: G06V 40/193 Preprocessing; Feature extraction

Abstract

Disclosed are methods, devices, and computer program products for red-eye detection in a digital image. In one example embodiment, a method for detecting a red-eye effect in a digital image includes several acts. First, red pixels having a predetermined degree of redness are identified in the image. Next, redness contrast is detected with respect to each of the red pixels and redness is then enhanced for those red pixels having a predetermined level of redness contrast. The pixels identified as being red are then further refined by applying another redness threshold based on one or more color characteristics associated with the red pixels. The refined set of red pixels may then be partitioned into a set of one or more candidate red-eye objects. A candidate red-eye object may be removed as a false positive based on geometric constraints associated with red-eye objects and/or proximity of the object to pixels with human skin-like color tones.

Claims

1. A method for red-eye detection in an input image, the method comprising the acts of:

enhancing redness of input image pixels having a threshold level of redness contrast; and

identifying candidate red-eye pixels in the enhanced image.

2. The method of claim 1, further comprising the acts of:

forming a candidate red-eye pixel map by designating pixels in the enhanced image as candidate red-eye pixels if the pixels have a first threshold level of redness; and

using the candidate red-eye pixel map to identify one or more candidate red-eye objects.

3. The method of claim 2, wherein a pixel in the enhanced image has the first threshold level of redness if a red component thereof exceeds each of a plurality of other color components thereof by a first set of respective threshold amounts.

4. The method of claim 3,

wherein the first set of threshold amounts are dependent on color characteristics of pixels in the input image that are identified as having a second threshold level of redness,

wherein a pixel in the input image has the second threshold level of redness if a red component thereof exceeds each of a plurality of other color components thereof.

5. The method of claim 4, wherein a pixel in the input image has the second threshold level of redness if a red component thereof exceeds each of a plurality of other color components thereof by a second set of respective threshold amounts.

6. The method of claim 2, further comprising the acts of:

evaluating one or more geometric characteristics of each of the one or more candidate red-eye objects that are indicative of whether each object is likely to be a red-eye object.

7. The method of claim 6, wherein evaluating the one or more geometric characteristics of a candidate red-eye object includes determining whether the candidate red-eye object has a substantially elongated shape indicative of a non-red-eye object.

8. The method of claim 6, wherein determining whether the candidate red-eye object has a substantially elongated shape indicative of a non-red-eye object includes determining if the object is shaped substantially like a straight line or if the object is substantially elongated in at least one of a horizontal, vertical, and diagonal orientation with respect to an orientation of the input image.

9. The method of claim 8, wherein determining if the object is substantially elongated in a diagonal orientation includes:

calculating a ratio between a total number of pixels in the object and an area of a bounding box for the object; and

comparing the ratio to a diagonal elongation threshold.

10. A method for red-eye detection in an input image, the method comprising the acts of:

generating a gamma-corrected image by performing image-adaptive gamma correction on at least a portion of the input image based on gray-scale intensity values of input image pixels; and

identifying an object in the input image as a candidate red-eye object based on whether, in the gamma-corrected image, pixels of a human skin-like color are detected proximate to the object in a predetermined number of directions.

11. The method of claim 10, wherein a pixel is determined to be skin-like if color components of the pixel are within predetermined limits.

12. The method of claim 11, wherein the color components of the pixel are expressed in another color space before determining whether the pixel is skin-like, and

wherein the predetermined limits are inclusive.

13. The method of claim 10, wherein the image-adaptive gamma correction is performed using a gamma that prevents unnecessary darkening of the image.

14. The method of claim 10, wherein the image-adaptive gamma correction is performed using a gamma derived from a mean gray-scale intensity value of the input image pixels and a mean gray-scale intensity value of a set of high intensity input image pixels.

15. The method of claim 14, wherein the set of high intensity input image pixels includes pixels having a gray-scale intensity value that is higher than a predetermined percentile of all gray-scale intensity values in at least a portion of the input image.

16. The method of claim 10, further comprising:

enhancing redness in pixels having a threshold level of redness contrast;

designating pixels in the enhanced image as candidate red-eye pixels if the pixels have a threshold level of redness; and

wherein the object identified as the candidate red-eye object is identified from among the candidate red-eye pixels.

17. The method of claim 10, further comprising:

evaluating one or more geometric characteristics of the candidate red-eye object that are indicative of whether the candidate red-eye object is likely to be a red-eye object.

18. A method for red-eye detection in an image, the method comprising the acts of:

identifying, in the image, a set of red pixels having a predetermined level of redness contrast with respect to neighboring pixels; and

partitioning pixels in the set of red pixels into a set of one or more candidate red-eye objects.

19. The method of claim 18, further comprising:

removing a pixel from the set of red pixels based on whether pixels of a human skin-like color are detected proximate to the pixel.

20. The method of claim 18, further comprising:

removing a candidate red-eye object from the set of one or more candidate red-eye objects based on whether pixels of a human skin-like color are detected proximate to the object.

24. An image processing device comprising:

a processor; and

the one or more computer-readable media as recited in claim 19, the computer-readable instructions thereon being configured to be executed by the processor.

21. One or more computer-readable media having computer-readable instructions thereon which, when executed, implement a method for processing an image to detect a red-eye effect, the method comprising the acts of:

identifying, in the image, a set of red pixels having a predetermined level of redness contrast with respect to neighboring pixels; and

partitioning pixels in the set of red pixels into a set of one or more candidate red-eye objects.

22. A method for red-eye detection in an image, the method comprising the acts of:

identifying, in the image, a first set of red pixels having a predetermined degree of redness;

detecting redness contrast in a plurality of directions with respect to each pixel in the first set of red pixels;

enhancing redness of pixels in the first set of red pixels that have a predetermined level of redness contrast in at least one of the plurality of directions;

identifying a second set of pixels from the first set of red pixels by deriving one or more color characteristics associated with individual pixels in the first set of red pixels and comparing the one or more color characteristics to one or more corresponding color characteristics common to all of the pixels in the first set of red pixels;

partitioning pixels in the second set of red pixels into a set of one or more candidate red-eye objects;

determining whether to remove a candidate red-eye object from the set of one or more candidate red-eye objects by applying geometric constraints to each of the one or more candidate red-eye objects;

performing image-adaptive gamma correction on the image using a gray-scaled version of the image; and

determining whether to remove a candidate red-eye object from the set of one or more candidate red-eye objects based on whether pixels with human skin-like color tones are detected in the gamma-corrected image in a predetermined number of directions proximate to each of the one or more candidate red-eye objects.

23. One or more computer-readable media having computer-readable instructions thereon which, when executed, implement a method for processing an image to detect a red-eye effect, the method comprising the acts of:

identifying, in the image, a first set of red pixels having a predetermined degree of redness;

detecting redness contrast in a plurality of directions with respect to each pixel in the first set of red pixels;

enhancing redness of pixels in the first set of red pixels that have a predetermined level of redness contrast in at least one of the plurality of directions;

identifying a second set of pixels from the first set of red pixels by deriving one or more color characteristics associated with individual pixels in the first set of red pixels and comparing the one or more color characteristics to one or more corresponding color characteristics common to all of the pixels in the first set of red pixels;

partitioning pixels in the second set of red pixels into a set of one or more candidate red-eye objects;

removing a candidate red-eye object from the set of one or more candidate red-eye objects by applying geometric constraints to each of the one or more candidate red-eye objects;

performing image-adaptive gamma correction on the image using a gray-scale histogram of the image; and

removing a candidate red-eye object from the set of one or more candidate red-eye objects based on whether pixels with human skin-like color tones are detected in the gamma-corrected image in a predetermined number of directions proximate to each of the one or more candidate red-eye objects.