Eye Feature Point Detection Using Edge Window Scanning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eye detection techniques struggle to accurately detect face feature points in face images due to the influence of ambient light and individual differences in facial structure, leading to inaccurate detection of features such as eyes, particularly when binarizing images.
Innovation Solution
The proposed eye detection apparatus employs a system that captures face images, sets an eye search region based on detected face features, calculates horizontal and vertical edge values using sobel filters, removes edges with subtle gray level differences, and uses an image window scanning method to determine the position of feature points with high accuracy, irrespective of ambient light and facial structure variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If binarized image processing is used for eye detection, then processing speed is improved, but detection accuracy deteriorates due to lighting conditions and individual facial structure differences
Solution Approach 1:
The patent divides the face image into multiple regions of interest (eyes, eyebrows, nose, mouth) and processes each region separately using appropriate methods. This segmentation allows different processing strategies to be applied to different areas, maintaining both speed and accuracy.
Solution Approach 2:
The patent changes the processing parameters based on the specific region being analyzed. Instead of using a single binarization threshold for the entire image, different parameters and methods are applied to different facial regions, adapting to local characteristics and improving detection accuracy while maintaining efficiency.
2Productivity
If extreme point-based detection is used to find eye candidates, then detection speed is improved, but detection accuracy deteriorates due to false positives from moles and other facial features
Solution Approach 1:
The patent implements a feedback mechanism where detection results from one method are used to guide and refine subsequent detection. The extreme point detection results are fed back into the region-based analysis, allowing false positives to be filtered out and true positives to be confirmed through multiple verification steps.
Solution Approach 2:
The patent introduces an intermediary verification process between initial candidate detection and final confirmation. Additional facial features (eyebrows, nose, mouth) serve as intermediary indicators to verify whether extreme points truly represent eyes, filtering out false positives from moles or other features.
3Productivity
If mask processing and labeling are performed on predetermined regions, then processing efficiency is improved, but detection accuracy deteriorates when feature points are not clearly separated from background
Solution Approach 1:
The patent applies local quality by using different processing approaches for different regions. Instead of uniform mask processing, the system adapts the processing method to local facial characteristics, using edge detection, gradient analysis, and region-specific thresholds to maintain accuracy in challenging areas while preserving processing efficiency.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables precise detection of eye feature points, including eyelids, with improved robustness against lighting conditions and individual facial differences, enhancing the accuracy of eye detection in face images.
Implementation Method 1
calculates horizontal and vertical edge values using sobel filters
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A face feature point detection apparatus (1) includes an image capturing device (2), an edge calculating unit (23) calculating edge values indicating a luminance change in a direction, and a detection target determining unit (14) scanning an edge image, which is created by arranging the edge values for corresponding pixels based on pixel arrangement of the face image, with an image window being an aggregation of selected pixels formed in a predetermined shape, the detection target determining unit (14) determining a position of the image window having a largest weighted sum of weighted sums to be a detection position where a detection target is present, providing that the weighted sum is calculated by multiplying the edge value which corresponds to each pixel in the image window by a predetermined value defined on a per-pixel basis and adding up all products of the edge value and the predetertxiined value.