| 摘要 |
Methods and apparatuses for detecting and correcting red-eye in an image are provided. In one aspect, red-eye can be detected without detecting faces. Red-eye candidates are identified and eliminated by evaluating a series of statistics, including color, size, shape, distance, etc. A plurality of regions, including an eye region, a pupil region, an iris region, and a sclera region can be associated with each red-eye candidate. Color distribution statistics are computed with respect to the presence of target color, glint, skin, or other-color pixels in the plurality of regions. Red-eye candidates may be paired with other red-eye candidates for further processing. Further statistics may be computed based on the location of the red-eye candidates in the image, and the location of each red-eye candidate relative to other red-eye candidates. Pixels associated with selected red-eye candidates may be color adjusted.
|
