Face Recognition Cluster Expansion for Age Variation Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face recognition technologies face challenges in reliably detecting and recognizing faces due to changes caused by factors like lighting, facial expressions, and pose, especially when insufficient training data is available, leading to unsatisfactory performance in real-world applications.
Innovation Solution
A method for face recognition using an intelligent electronic device that involves obtaining face information, comparing it to a default cluster, extracting a face vector if necessary, determining the age variation state using an artificial neural network, and configuring an expanded cluster by adding the face feature vector to the default cluster, allowing for improved recognition across age-related changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If MLP-based learning methods are used to minimize errors from given sample data, then the method works well on learned data, but it does not ensure successful operation on new data which has not yet learned
Solution Approach 1:
The patent applies preliminary action by pre-processing face images through alignment and normalization before feeding them to the neural network. Face images are aligned based on detected key points (eyes, nose, mouth) and normalized to a standard size, ensuring that the network receives consistently formatted input data. This preliminary preparation improves the network's ability to generalize to new faces while maintaining reliability on training data.
Solution Approach 2:
The patent employs parameter changes by dynamically adjusting neural network parameters including learning rate, batch size, and network depth based on training progress and data characteristics. The system also transforms face images into various parameter spaces (aligned coordinates, normalized pixel values, extracted features) to optimize training convergence and improve generalization performance on unseen faces.
2Reliability
If a vast amount of sample data is provided to MLP-based methods, then reliability improves, but the complexity and resource requirements increase significantly
Solution Approach 1:
The patent extracts essential features from face images using detected key points (eyes, nose, mouth, eyebrows) and their geometric relationships. Instead of processing entire high-dimensional images, the system extracts and utilizes only the critical facial landmarks and their relative positions, significantly reducing data complexity while maintaining recognition reliability. This extraction approach allows effective training with fewer samples.
Solution Approach 2:
The patent segments the face recognition task into distinct stages: face detection, key point detection, alignment transformation, feature extraction, and classification. This segmentation allows each component to be optimized independently and enables the system to handle complex recognition tasks through a series of simpler, more manageable operations, reducing overall system complexity.
3Reliability
If nonlinear SVM is used to obtain reliable face detection performance, then detection reliability improves, but the execution speed of the algorithm becomes slow due to a large amount of computation
Solution Approach 1:
The patent replaces the traditional mechanical SVM classification approach with a deep neural network that learns decision boundaries automatically through gradient-based optimization. This substitution allows the system to achieve comparable or superior reliability while enabling faster execution through optimized neural network inference and hardware acceleration, eliminating the computational bottleneck of conventional SVM.
Solution Approach 2:
The patent applies preliminary action by pre-aligning and normalizing face images before classification, and by pre-training the neural network on a large dataset to learn robust feature representations. This preliminary preparation reduces the computational complexity of the actual detection phase, enabling faster execution while maintaining high reliability on unseen faces.
Data Source
AI summary
According to an embodiment, a method for recognizing a user's face using an intelligent electronic device comprises obtaining a face area from the user's face captured and obtaining face information from the face area, comparing the obtained face information with a default cluster, selecting whether to extract a face vector from the face information according to a result of the comparison, determining an age variation state for the user's face based on the extracted face vector, and upon determining that the face vector is in the age variation state for the user's face, extracting a face feature vector from the face vector and configuring an expanded cluster by adding the face feature vector to the default cluster. According to the disclosure, the intelligent electronic device may be related to artificial intelligence (AI) modules, unmanned aerial vehicles (UAVs), robots, augmented reality (AR) devices, virtual reality (VR) devices, and 5G service-related devices.