Face Recognition via Robust Collaborative Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face recognition techniques face challenges in accurately identifying faces under complex occlusions and corruptions, particularly when training images are not carefully controlled and the number of samples per class is insufficient, and existing methods are inefficient in removing corruptions from new data.
Innovation Solution
A face recognition method and system that employs a robust collaborative representation-based classifier using inductive robust principal component analysis (IRPCA) and l1 half quadratic (LRR-HQ-L1) algorithms to recover clean images from occlusions and corruptions, combined with a dictionary for storing face images and a face classifier for classification, enabling efficient recognition in video frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If l1 optimization methods are used for face recognition, then face recognition accuracy is improved under controlled conditions, but the method fails when training images are not carefully controlled or sample size is insufficient
Solution Approach 1:
The patent applies Robust Principal Component Analysis (RPCA) to pre-process training images and remove outlier pixels before the main face recognition process. This preliminary cleaning action ensures that subsequent l1 optimization methods work with clean data, making the system robust to uncontrolled conditions and insufficient sample sizes.
Solution Approach 2:
The patent introduces a two-stage process where RPCA acts as an intermediary between corrupted training data and the face recognition algorithm. The RPCA component separates low-rank clean images from sparse corruption, providing clean input to the recognition system without requiring controlled conditions or large sample sizes.
2Reliability
If RPCA and LRR methods are used to remove corruptions from training data, then clean images are recovered, but these methods are transductive and cannot efficiently remove corruptions from new test data
Solution Approach 1:
The patent transforms the static, transductive RPCA approach into a dynamic, inductive system. The method learns the low-rank structure from training data and applies this learned structure to new test samples, enabling the system to adapt to unseen data while maintaining the ability to recover clean images from corruptions.
Solution Approach 2:
The patent performs preliminary low-rank representation learning on training data to establish a clean subspace structure. This preliminary action creates a robust framework that can then be applied inductively to new test samples, enabling corruption removal without re-processing the entire training set.
3Measurement precision
If face recognition systems process videos with complex occlusions and corruptions, then recognition accuracy improves with more sophisticated methods, but computational complexity and processing time increase
Solution Approach 1:
The patent segments the face recognition process into distinct modules: RPCA for corruption removal, LRR for low-rank structure learning, and final classification. This segmentation allows each component to be optimized independently and processed efficiently, reducing overall computational complexity while maintaining high accuracy under occlusion.
Data Source
AI summary
A face recognition method is provided. The method includes dividing an input video into different sets of frames and detecting faces of each frame in the input video. The method also includes generating face tracks for the whole video. Further, the method includes applying a robust collaborative representation-based classifier to recover a clean image from complex occlusions and corruptions for a face test sample and perform classification. In addition, the method also includes outputting the video containing the recognized face images.