Red-eye removal using multiple recognition channels

摘要

This disclosure pertains to apparatuses, methods, and computer readable media for red-eye removal techniques using multiple recognition channels. In the following examples, red, golden, and white recognition channels are used. A recognition channel is the monochrome extraction from a color photograph in a manner designed to make one kind of red-eye artifact glow with maximum contrast. Once the red-eye artifact has been characterized by, e.g., size and location, the techniques disclosed herein may then discern whether the red-eye artifact is, for example, a red-, golden-, or white-eye case by examining the configuration and characteristics of prominence bitmasks created for the various recognition channels. Once the type of red-eye case has been discerned, the techniques disclosed herein may then replace the artifact with a photographically reasonable result based on the type of red-eye case being repaired. Specular reflection may also be re-added to the photograph.

主权项

1. A processor-implemented image processing method, comprising:
receiving, at the processor, a plurality of prominence bitmasks, wherein each prominence bitmask indicates an amount of color prominence of a respective color in a color image stored in a memory and the prominence bitmasks include pixels corresponding to an artifact common to each of the plurality of prominence bitmasks; receiving, at the processor, a plurality of metric values for each of the prominence bitmasks; determining, by the processor, a plurality of topological configurations of the plurality of prominence bitmasks, wherein the topological configurations comprise an overlap of the plurality of prominence bitmasks; and identifying, by the processor, the common artifact as a artifact type classified by color, wherein the identification is based at least in part on one or more of the plurality of metric values and at least one of the plurality of topological configurations.