Automatic red eye detection and correction in digital images
First Claim
1. A method for performing red eye correction in an image, comprising:
- providing a digital image, the digital image being represented by electrical signals in a computer system;
identifying a red eye region in the digital image; and
applying a color correction to each pixel in the identified red eye region; and
wherein the identifying of the red eye region comprises;
identifying hard red areas, each of the hard red areas comprising a grouping of pixels having a color defined within a polyhedral region of a three dimensional hue-saturation-intensity (HSI) color space having a first axis defining hue, a second axis defining intensity, and a third axis defining saturation; and
excluding ones of the hard red areas from being identified as the red eye region when a virtual weight center of the hard red area deviates more than a selected percentage from a geometric center of the hard red area, the virtual weight center having an x-coordinate equal to a sum of an x-coordinate of each pixel of the hard red area divided by a number of pixels in the hard red area, and a y-coordinate equal to a sum of a y-coordinate of each of the pixels within the hard red area divided by the number of pixels in the hard red area.
13 Assignments
0 Petitions
Accused Products
Abstract
A method for the automatic detection and correction of red eye in a digital image is disclosed. The method includes defining a digital image in a hue-saturation-intensity (HSI) color space, and identifying a red eye region in a digital image. Using HSI criteria, identified regions are filtered to discard areas unlikely to be the result of red eye effect, and then a plurality of algorithms are used to apply a color correction to each pixel in the identified red eye region. The color correction manipulates each pixel of the red eye region remove the red eye effect. The method is automatic, and requires no input from a user to define the red eye region, to identify the true color of the red eye region, or to apply the color correction.
85 Citations
12 Claims
-
1. A method for performing red eye correction in an image, comprising:
-
providing a digital image, the digital image being represented by electrical signals in a computer system; identifying a red eye region in the digital image; and applying a color correction to each pixel in the identified red eye region; and wherein the identifying of the red eye region comprises; identifying hard red areas, each of the hard red areas comprising a grouping of pixels having a color defined within a polyhedral region of a three dimensional hue-saturation-intensity (HSI) color space having a first axis defining hue, a second axis defining intensity, and a third axis defining saturation; and excluding ones of the hard red areas from being identified as the red eye region when a virtual weight center of the hard red area deviates more than a selected percentage from a geometric center of the hard red area, the virtual weight center having an x-coordinate equal to a sum of an x-coordinate of each pixel of the hard red area divided by a number of pixels in the hard red area, and a y-coordinate equal to a sum of a y-coordinate of each of the pixels within the hard red area divided by the number of pixels in the hard red area. - View Dependent Claims (2, 3, 4, 5)
-
-
6. A method for performing red eye correction in a digital image, comprising:
-
identifying hard red areas in the digital image, each of the hard red areas comprising a grouping of pixels having a color defined within a polyhedral region of a three dimensional hue-saturation-intensity (HSI) color space having a first axis defining hue, a second axis defining intensity, and a third axis defining saturation; identifying a real color center for the identified hard red areas, the real color center being defined as at least one pixel defining a point within the identified hard red area, the at least one pixel having values for hue, saturation and intensity closest of all pixels within the hard red area to the arithmetic averages of all pixels within the hard red area for hue, saturation, and intensity; identifying a set of final red areas, the set of final red areas including the hard red areas and same color areas, each of the same color areas being a contiguous set of pixels built around the real color centers of the hard red areas wherein each pixel of the contiguous set of pixels has a color that is within a predetermined color distance from an average color of the hard red area; filtering out non-red-eye red areas from the set of final red areas, the filtering including the disregarding of ones of the final red having areas that are too large for red eye effect, areas of an inappropriate shape to have red eye;
areas of insufficient color intensity;
areas of insufficient brightness dispersion, and areas failing to match a pre-determined spectral criteria;defining a region having red eye effect, the region having red eye effect being one of the final red areas that survived the filtering; and applying a color correction to the defined region, wherein the method is performed automatically and without user input to define and to correct the region having red eye effect. - View Dependent Claims (7, 8)
-
-
9. Computer readable media encoded with a computer program for removal of red eye effect in a digital image, the computer program comprising:
-
program instructions for automatically defining a region of the digital image having red eye effect; and program instructions for automatically applying a color correction to the defined region, wherein the defining of the region of the digital image having the red eye effect comprises; identifying hard red areas, each of the hard red areas comprising a grouping of pixels having a color defined within a polyhedral region of a three dimensional hue-saturation-intensity (HSI) color space having a first axis defining hue, a second axis defining intensity, and a third axis defining saturation; and excluding ones of the hard red areas from being identified as the red eye region when a virtual weight center of the hard red area deviates more than a selected percentage from a geometric center of the hard red area, the virtual weight center having an x-coordinate equal to a sum of an x-coordinate of each pixel of the hard red area divided by a number of pixels in the hard red area, and a y-coordinate equal to a sum of a y-coordinate of each of the pixels within the hard red area divided by the number of pixels in the hard red area. - View Dependent Claims (10, 11, 12)
-
Specification