Automatic Redeye Reduction

Redeye is a common problem in consumer photography. It occurs when a flash is used in photography and the light reflecting from human retina makes the eyes appear red. This artifact is drawing increased attention with the popularity of digital cameras, in which camera size continues to shrink and the distance between the flash and the lens is decreasing.  Although some cameras are equipped with pre-flash hardware to alleviate redeye problems, they are not well received since they do not always prevent the redeye artifact and tend to result in people blinking during the actual photo taking. At HP Labs, we are working on digital solutions to remove red eyes in the post-processing stage.

Currently, many image processing software applications in the market offer redeye removal solutions. However, the majority of them are semi-automatic or manual solutions. The user has to either click on the redeye or draw a box containing the redeye before the redeye removal algorithms can find the redeye pixels and correct them. Besides inconvenience, a typical problem with these algorithms is poor segmentation. Even with user input, these algorithms sometimes identify redeye pixels too aggressively, darkening eye lid areas, or too conservatively, leaving many redeye pixels uncorrected.

In contrast, HP solution comprises of a fully automatic redeye detection module and a color correction module, which finds and corrects red eyes without user interventions.

In the past years, researchers at HPL have built up core expertise in two important areas that are relevant to the redeye problem: pattern recognition and color science. Our most recent red eye solution includes the following algorithm design:

• A training based detector design, which uses Adaboost algorithm to simultaneously select relevant features and assign feature weight for the classifier.
• A systematic, orientation independent feature computation scheme for object detection.Because of this design, our trained detector is able to detect red eyes of arbitrary in-plane rotation.
• An adaptive, high quality red eye recoloring algorithm.

In addition, because of the high efficiency of the algorithm, we have successfully implemented in various embedded platforms in HP products.

For more technical details, please refer to the following papers:

H. Luo, J. Yen and D. Tretter, "An Efficient Automatic Redeye Detection and Correction Algorithm”, to appear in ICPR 2004. (pdf download)

A. Patti, K. Konstantinides, D. Tretter and Q. Lin, "Automatic Digital Redeye Detection", ICIP, 1999.

The most recent red eye remove algorithm we developed have been used in various HP products. Among them includes: HP Photosmart R707 camera, HP Photosmart 7960 printer , etc. In addition, this algorithm is also included in HP's RedBot web service.

For more information about HPL redeye technology, please contact Huitao Luo (huitao.luo@hp.com) at Imaging Technology Dept., HP Labs.