Pose-Independent Face Demographics Classification Using Parallel Regressors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face-based demographics classification systems are not robust to pose changes and accidental scene variables like noise, lighting, and occlusion, leading to inaccurate tracking and classification.
Innovation Solution
A method and system that estimates and corrects two-dimensional and three-dimensional facial poses using a parallel array of multiple learning machine regressors, builds pose-dependent appearance models, and performs demographics classification only once per track, utilizing filtered representations and edge responses to suppress irrelevant variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional face-based demographics classification systems are used, then the system is simple to implement, but the system is not robust to pose changes and scene variables leading to inaccurate tracking and classification
Solution Approach 1:
The patent segments the face classification problem by creating separate pose-dependent appearance models for different facial poses (frontal, profile, etc.). Each model is trained independently on faces with specific pose characteristics, allowing the system to select the appropriate model based on the detected pose. This segmentation improves robustness to pose changes while keeping each individual model relatively simple.
Solution Approach 2:
The patent changes the parameter space by estimating and utilizing three-dimensional pose parameters (yaw, pitch, roll) of faces. By incorporating pose estimation as an additional parameter, the system can adapt its classification approach based on the detected pose, significantly improving robustness to pose variations and scene variables.
2Measurement precision
If pose-dependent facial models are built for each pose, then the classification accuracy improves, but the computational resources and processing time increase
Solution Approach 1:
The patent performs preliminary pose estimation and face detection before classification. By estimating the three-dimensional pose early in the processing pipeline and using it to select the appropriate pose-dependent model, the system avoids unnecessary computation. Only the relevant pose-specific model is applied, maintaining high accuracy while improving processing efficiency through selective computation.
Solution Approach 2:
The patent implements a dynamic model selection mechanism that adapts to the input face pose. Based on the estimated pose parameters, the system dynamically selects which pose-dependent appearance model to apply for classification. This dynamic approach ensures high accuracy for each pose type while avoiding the computational overhead of applying all models to every face.
3Adaptability or versatility
If multiple pose-dependent appearance models are maintained, then the system becomes more adaptable to different poses, but the device complexity and memory requirements increase
Solution Approach 1:
The patent segments the pose space into discrete categories (frontal, profile, and intermediate poses) and creates dedicated appearance models for each segment. This segmentation provides adaptability to different poses while limiting the total number of models to a manageable set, balancing versatility with system complexity.
Solution Approach 2:
The patent creates a universal framework that handles multiple pose types through a common architecture. The pose estimation module and model selection mechanism serve multiple functions: detecting pose, categorizing it, and selecting the appropriate model. This multi-functional approach increases adaptability without proportionally increasing overall system complexity.
Data Source
AI summary
The invention provides a face-based automatic demographics classification system that is robust to pose changes of the target faces and to accidental scene variables, by using a pose-independent facial image representation which comprises multiple pose-dependent facial appearance models. Given a sequence of people's faces in a scene, the two-dimensional variations are estimated and corrected using a novel machine learning based method. We estimate the three-dimensional pose of the people, using a machine learning based approach. The face tracking module keeps the identity of the person using geometric and appearance cues, where multiple appearance models are built based on the poses of the faces. Each separately built pose-dependent facial appearance model is fed to the demographics classifier, which is trained using only the faces having the corresponding pose. The classification scores from the set of pose-dependent classifiers are aggregated to determine the final face category, such as gender, age, and ethnicity.