Pupil and Iris Detection Using Gradient Sign Likelihood
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing devices face challenges in accurately detecting the pupil or iris in face images due to issues like edge detection errors from ambient light and hardware requirements for high processing capacity, especially in environments with reflections and moving eyes.
Innovation Solution
An image processing device that generates an edge image with gradient magnitude and sign information, uses a sampling curve projected on a three-dimensional model to calculate likelihood, and detects the pupil or iris by distinguishing between positive and negative gradients, thereby improving detection accuracy and reducing noise from ambient features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the Hough transform is used for high accuracy detection, then detection precision is improved, but device complexity and processing capacity requirements increase
Solution Approach 1:
The patent segments the detection process into multiple stages: initial edge detection using Sobel filter, candidate region extraction based on geometric constraints, and final verification using simplified Hough transform only on candidates. This segmentation reduces the overall computational load while maintaining detection accuracy.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. Full Hough transform is applied only to candidate regions identified by preliminary processing, rather than the entire image. This local application of complex processing maintains accuracy where needed while reducing overall complexity.
2Loss of information
If edge detection is performed unconditionally on all objects in the image, then comprehensive feature extraction is achieved, but detection accuracy deteriorates due to noise from non-target objects
Solution Approach 1:
The patent extracts and isolates the eye region from the rest of the face image before performing edge detection. By cropping the image to contain only the eye region and applying geometric constraints (aspect ratio, size ranges), the method extracts only relevant features while excluding noise from other facial features like eyebrows and nose.
Solution Approach 2:
The patent introduces an intermediary processing stage between initial edge detection and final detection. This intermediary stage includes gradient magnitude thresholding and candidate region filtering, which acts as a mediator to separate target-related edges from noise edges before final pupil/iris detection.
3Productivity
If binarization or ternarization is performed during image processing, then processing speed is improved, but detection accuracy is affected by ambient light conditions
Solution Approach 1:
The patent performs preliminary gradient calculation and threshold determination before binarization. By pre-calculating the Sobel gradients and determining appropriate thresholds based on image statistics, the method prepares the data in advance to make binarization more robust against ambient light variations, thus maintaining both speed and accuracy.
4Measurement precision
If ellipse detection algorithm is used, then circular shapes are detected accurately, but detection fails when the eye moves sideways and the shape collapses
Solution Approach 1:
The patent dynamically adjusts the detection model from a fixed ellipse to a more flexible representation that can accommodate eye movements. By using gradient-based edge detection and allowing the detection window to shift and adapt to the actual eye position, the method maintains accuracy even when the eye moves sideways and the pupil/iris shape appears distorted in the image.
Data Source
AI summary
An image processing device includes a data processing unit that processes data of a face image which is captured to include a face. The data processing unit generates an edge image by filtering the face image to detect an edge in a scanning direction, extracts a sampling value as the information regarding a gradient magnitude and whether the gradient is positive or negative from each of positions in the edge image corresponding to a plurality of points constituting the sampling curve, calculates a likelihood with respect to the sampling curve by setting points having a positive gradient and a negative gradient as likelihood evaluation targets in a first point group and a second point group, and detects a sampling curve having a maximum likelihood as a pupil or an iris among a plurality of sampling curves.