Automatic red-eye repair using multiple recognition channels

摘要

This disclosure pertains to apparatuses, methods, and computer readable media for automatic red-eye repair using multiple recognition channels. While it is possible to manually specify all of the eyes in an image to be repaired, it is desirable for repair to happen automatically. Since red-eye repair algorithms are dependent upon knowing the image position and size of each artifact to be repaired, in an automatic repair mode, the algorithm must be directed as to where the repair should be applied. Face detection is one way to determine eye positions and the interocular distance (IOD) with some degree of certainty. In some embodiments, red, golden, and white recognition channels may be used to locate and determine the type of the artifacts. Once an artifact has been characterized by, e.g., type, size, and location, the techniques disclosed herein may then repair the artifact, replacing it with a photographically reasonable result.

主权项

1. An automatic artifact repair method, comprising:
receiving face location information for a face in an image, the face location information comprising two eye points, the image stored in a memory; automatically identifying a candidate artifact for a first eye in the face based at least in part on evaluation of at least one of a plurality of recognition channels for the first eye, wherein the first eye is associated with a first of the two eye points, and wherein each recognition channel comprises a monochrome extraction from the image designed to make one kind of artifact exhibit maximum contrast; automatically generating a candidate repair for the candidate artifact based, at least in part, on the face location information, wherein the candidate repair comprises repair information for a first plurality of pixels comprising the identified candidate artifact for the first eye; automatically determining a confidence measure for the candidate repair based, at least in part, on at least one of the following: a size of the candidate repair, a strength of the candidate repair, a contrast of the candidate repair, and the face location information; automatically applying the candidate repair to the image stored in the memory if the confidence measure is greater than a threshold value; and automatically rejecting the candidate repair if the confidence measure is less than the threshold value.