Subspace Feature Matching for Efficient Facial Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial recognition techniques using convolutional neural networks (CNNs) are computationally complex, requiring significant memory, processing power, and time, which is a challenge for portable devices that need efficient object recognition.
Innovation Solution
The method involves generating a subspace spanned by multiple representative feature vectors, where a probe feature vector is compared to a reconstructed feature vector within this subspace to facilitate efficient object recognition, using techniques such as convolutional operations and linear combinations of weighting coefficients.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If CNN-based facial recognition is used, then recognition accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent extracts the essential information from the full CNN feature vectors by projecting them onto a lower-dimensional subspace spanned by representative feature vectors. This extraction process retains the most discriminative features while discarding redundant information, thereby reducing computational complexity while maintaining recognition accuracy.
Solution Approach 2:
The patent segments the high-dimensional feature space into a manageable subspace by selecting a subset of representative feature vectors that span the essential variation in the data. This segmentation allows the system to perform recognition operations in a reduced-dimensional space, reducing the computational burden on portable devices.
2Measurement precision
If CNN-based facial recognition is used, then recognition accuracy is improved, but memory requirements and power consumption increase
Solution Approach 1:
The patent extracts and stores only the essential subspace information represented by a compact set of representative feature vectors, rather than storing and processing complete high-dimensional CNN feature vectors. This extraction significantly reduces memory requirements and subsequent power consumption during recognition operations on portable devices.
Solution Approach 2:
The patent performs preliminary action by pre-computing and storing the subspace basis (representative feature vectors) before actual recognition occurs. This preliminary processing allows the recognition system to operate with reduced computational complexity during real-time operations, thereby reducing power consumption on portable devices.
3Reliability
If full CNN feature vectors are processed, then recognition robustness is improved, but processing time increases
Solution Approach 1:
The patent segments the feature representation into a compressed form by projecting full CNN feature vectors onto a lower-dimensional subspace. This segmentation maintains the essential robustness needed for accurate recognition while significantly reducing the number of computations required during processing, thereby reducing processing time.
Solution Approach 2:
The patent introduces an intermediary subspace representation that mediates between the full CNN feature vectors and the final recognition decision. This intermediary subspace projection acts as a computational bridge that preserves recognition robustness while reducing the time required for processing operations on portable devices.
Data Source
AI summary
Methods, systems, and devices for object recognition are described. A device may generate a subspace based at least in part on a set of representative feature vectors for an object. The device may obtain an array of pixels representing an image. The device may determine a probe feature vector for the image by applying a convolutional operation to the array of pixels. The device may create a reconstructed feature vector in the subspace based at least in part on the set of representative feature vectors and the probe feature vector. The device may compare the reconstructed feature vector and the probe feature vector and recognize the object in the image based at least in part on the comparison. For example, the described techniques may support pose invariant facial recognition or other such object recognition applications.