System and method for automatically detecting and correcting red eye
First Claim
1. A system for detecting and correcting red-eye in an image without user intervention, the system comprising:
- a face recognition and location engine configured to identify one or more faces within the image, wherein at least a portion of the image is in the chrominance-luminance color space;
a red-eye locator engine configured to detect an occurrence of red-eye within the one or more identified faces, wherein the red-eye locator engine computes a color variation metric value for a given pixel of an identified face by computing a mean luminance-chrominance value for a patch of pixels centered on the given pixel, a squared distance between the luminance-chrominance value of the given pixel and the computed mean luminance-chrominance value for the patch, and an average of the squared distances within a region centered on the given pixel; and
a red-eye correction engine configured to correct the detected occurrences of red-eye within the image, wherein the red-eye locator engine detects an occurrence of red-eye based upon one or more metrics computed for one or more identified faces, the one or more metrics including a color variation metric tat corresponds to a change in color within a given area, the color variation metric corresponds to a change in both luminance and chrominance within the given area.
3 Assignments
0 Petitions
Accused Products
Abstract
A system and method automatically detects and corrects the occurrence of red-eye in digital photographs and images without user intervention. The system includes a face recognition and locator engine for locating human faces within the image and for generating a bounding box around each face. A red-eye locator engine analyzes the pixel data for each bounding box and computes one or more predefined metrics. The preferred metrics include color variation, redness, redness variation and glint. The red-eye locator engine also generates one or more detection masks based upon the computed metrics, and searches the detection mask for an occurrence of red-eye. A red-eye correction engine receives the detection mask including the detected occurrences of red-eye, and generates a correction mask. Pixels identified as being occurrences of red-eye are then de-saturated in accordance with the correction mask.
-
Citations
20 Claims
-
1. A system for detecting and correcting red-eye in an image without user intervention, the system comprising:
-
a face recognition and location engine configured to identify one or more faces within the image, wherein at least a portion of the image is in the chrominance-luminance color space; a red-eye locator engine configured to detect an occurrence of red-eye within the one or more identified faces, wherein the red-eye locator engine computes a color variation metric value for a given pixel of an identified face by computing a mean luminance-chrominance value for a patch of pixels centered on the given pixel, a squared distance between the luminance-chrominance value of the given pixel and the computed mean luminance-chrominance value for the patch, and an average of the squared distances within a region centered on the given pixel; and a red-eye correction engine configured to correct the detected occurrences of red-eye within the image, wherein the red-eye locator engine detects an occurrence of red-eye based upon one or more metrics computed for one or more identified faces, the one or more metrics including a color variation metric tat corresponds to a change in color within a given area, the color variation metric corresponds to a change in both luminance and chrominance within the given area. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-
-
17. A method for detecting and correcting red-eye in an image without user intervention, the method comprising the steps of:
-
identifying one or more faces within the image, wherein at least a portion of the one or more faces are represented by pixels in luminance-chrominance color space; detecting occurrences of red-eye for an identified face based on one or more metrics computed for at least some of the pixels of the identified face, wherein the detecting comprises the steps of;
computing a mean luminance-chrominance value for a patch pixels centered on a given pixel, computing a squared distance between the luminance-chrominance value of the given pixel and the computed mean luminance-chrominance value for the patch, computing an average of the squared distance within a region centered on the given pixel, and assigning the computed average as the color variation metric value for the given pixel; andcorrecting the detected occurrences of red-eye, wherein the one or more metrics includes a color variation metric that corresponds to a change in both luminance and chrominance within a given area of the identified face. - View Dependent Claims (18, 19, 20)
-
Specification