Pseudo 2D HMM Face Detection for Iris Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face detection methods are inadequate for identifying faces in complex backgrounds and multiple occurrences, especially in crowded scenes, and fail to accurately separate faces from non-faces, which hinders efficient iris recognition processes.
Innovation Solution
The apparatus employs a Pseudo 2D Hidden Markov Model (HMM) to detect and localize facial features by training on statistical feature vectors from face images, using a directed acyclic graph (DAG) to search for faces in images, and extracts observation vectors from 2D-DCT coefficients to identify facial structures like forehead, eyes, nose, and mouth, even in the presence of occlusions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional face detection methods are used, then the system is simple to implement, but the detection accuracy deteriorates in complex backgrounds with multiple faces
Solution Approach 1:
The face detection system segments the face into multiple facial features (eyes, nose, mouth, forehead) and models each as separate HMM states. This segmentation allows the system to detect faces more accurately by verifying the presence and spatial relationships of multiple features rather than relying on a single holistic face template, thereby improving detection precision in complex backgrounds.
Solution Approach 2:
The patent transitions from traditional 2D face detection to a pseudo-3D representation by stacking multiple 2D face images along the time dimension to create training data for HMM. This dimensional transformation enables the model to capture temporal variations and spatial relationships across multiple facial features, improving detection accuracy while maintaining computational feasibility.
2Measurement precision
If face detection is performed in crowded scenes with multiple faces, then complete face detection coverage is improved, but the ability to separate individual faces from background deteriorates
Solution Approach 1:
The HMM face model segments the face into distinct anatomical regions (eyes, nose, mouth, forehead) with specific spatial relationships. This segmentation enables the system to distinguish individual faces from backgrounds and other faces by verifying the characteristic spatial arrangement of facial features, thereby improving face separation accuracy in crowded scenes.
Solution Approach 2:
The patent introduces statistical feature vectors extracted from multiple facial images as an intermediary representation between the raw image data and the HMM model. These feature vectors capture the essential characteristics of facial structures and serve as a bridge that simplifies the detection process while maintaining high accuracy in separating faces from complex backgrounds.
3Adaptability or versatility
If statistical feature vectors from multiple images are used for training, then the robustness to facial variations improves, but the training time and computational resources increase
Solution Approach 1:
The patent extracts only the most relevant statistical feature vectors from multiple facial images for HMM training, rather than using all available image data. This selective feature extraction approach captures the essential variations in facial structures (eyes, nose, mouth, forehead) while reducing the computational burden and training time, achieving a balance between robustness and efficiency.
4Measurement precision
If the face is divided into sequential stripes for HMM training, then the localization of facial features improves, but the model complexity increases
Solution Approach 1:
The patent divides the face into sequential horizontal stripes or regions (forehead, eyes, nose, mouth, chin) and assigns each region as a separate HMM state. This segmentation enables precise localization of facial features by tracking the spatial position and characteristics of each segmented region, improving feature localization precision while the sequential structure keeps the model complexity manageable.
Data Source
AI summary
An apparatus for automatic detection of the face in a given image and localization of the eye region which is a target for recognizing iris is provided. The apparatus includes an image capturing unit collecting an image of a user; and a control unit extracting a characteristic vector from the image of the user, fitting an extracted vector into a Pseudo 2D Hidden Markov Model (HMM), and an operating method thereof for detecting a face and facial features of the user.