Red Eye Detection Using Face Orientation and Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic redeye reduction systems in images have high computation and memory requirements, struggle with detecting non-frontal faces, and often incorrectly identify regions as redeye defects, leading to false positives and visually displeasing corrections.
Innovation Solution
A method that performs initial segmentation of candidate redeye regions, uses face location and orientation information to determine the probability of redeye regions, and removes false positives before correction, ensuring accurate redeye detection and reduction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face detection algorithms are used to detect redeye regions, then redeye detection accuracy is improved, but computation and memory resource requirements increase
Solution Approach 1:
The patent segments the redeye detection process into multiple stages: initial candidate region identification using color and shape analysis, face detection to verify eye locations, and final confirmation. This multi-stage segmentation allows the system to use complex face detection only for verifying candidates rather than analyzing every pixel, thus improving accuracy while controlling computational resources.
Solution Approach 2:
The patent performs preliminary filtering of candidate redeye regions based on color properties and shape characteristics before applying computationally intensive face detection algorithms. By pre-processing and filtering candidates, the system reduces the number of regions requiring full face detection analysis, thereby lowering overall computation and memory requirements while maintaining detection accuracy.
2Adaptability or versatility
If traditional face detection algorithms are used, then frontal faces can be detected, but rotated faces (in-plane or out-of-plane) cannot be detected
Solution Approach 1:
The patent employs dynamic face detection capabilities that can adapt to various face orientations including rotated in-plane and out-of-plane faces. The system uses multiple detection passes and adjusts detection parameters based on detected face orientations, enabling reliable detection across different pose conditions rather than being limited to fixed frontal views.
3Manufacturing precision
If red eye correction is applied to all detected red regions, then redeye defects are corrected, but false positive regions are also incorrectly modified
Solution Approach 1:
The patent implements a feedback mechanism where detected candidate regions are verified against face detection results and orientation information before correction is applied. The system uses multiple verification stages that provide feedback loops to confirm whether a red region is truly an eye, allowing the system to reject false positives while maintaining correction accuracy for actual redeye defects.
Solution Approach 2:
The patent performs preliminary verification of candidate redeye regions using face detection and orientation analysis before applying correction. By pre-verifying that candidate regions actually correspond to eyes in rotated or non-frontal faces, the system prevents false positive corrections while ensuring accurate correction of genuine redeye defects.
Data Source
AI summary
An image is acquired including a red eye defect and non red eye defect regions having a red color. An initial segmentation of candidate redeye regions is performed. A location and orientation of one or more faces within the image are determined. The candidate redeye regions are analyzed based on the determined location and orientation of the one or more faces to determine a probability that each redeye region appears at a position of an eye. Any confirmed redeye regions having at least a certain threshold probability of being a false positive are removed as candidate redeye defect regions. The remaining redeye defect regions are corrected and a red eye corrected image is generated.