Face Recognition via Sparse Linear Combination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition systems are not robust to occlusions, variations in illumination, alignment, and pose, leading to decreased performance in real-world applications.
Innovation Solution
A system using l1-minimization to recover sparse coefficients from high-dimensional data, allowing for effective recognition of faces under occlusions by representing test images as a sparse linear combination of training images plus sparse error, and an alignment algorithm that iteratively minimizes registration errors to handle misalignments and varying illuminations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If classical recognition algorithms like nearest-neighbor are used, then the system is simple to implement, but recognition accuracy deteriorates under occlusion
Solution Approach 1:
The patent transforms the recognition problem from direct image comparison to sparse coefficient recovery in a transformed domain. By representing images as sparse linear combinations of training images and solving for coefficients using l1-minimization, the system achieves robustness to occlusion while maintaining computational efficiency through convex optimization.
2Reliability
If prior knowledge of occlusion location is required, then recognition accuracy improves, but the system becomes less adaptable to unexpected occlusions
Solution Approach 1:
The system automatically recovers the sparse occlusion pattern without requiring prior knowledge of occlusion location or type. The l1-minimization framework simultaneously solves for both the clean image representation and the sparse error term, allowing the algorithm to adapt to any occlusion pattern that satisfies the sparsity assumption.
3Reliability
If high-dimensional data is used for recognition, then recognition accuracy improves, but computational complexity increases
Solution Approach 1:
The patent extracts the essential information from high-dimensional images by representing them as sparse linear combinations of training images. This sparse representation dramatically reduces the effective dimensionality of the problem, allowing accurate recognition without processing the full high-dimensional data directly.
4Reliability
If alignment algorithms are added to handle misalignment, then recognition robustness improves, but system complexity increases
Solution Approach 1:
The patent merges the alignment and recognition tasks into a single unified sparse representation framework. By representing misaligned images as sparse combinations of training images with appropriate transformations, the system achieves alignment robustness without requiring separate alignment preprocessing steps.
Data Source
AI summary
A method is disclosed for recognition of high-dimensional data in the presence of occlusion, including: receiving a target data that includes an occlusion and is of an unknown class, wherein the target data includes a known object; sampling a plurality of training data files comprising a plurality of distinct classes of the same object as that of the target data; and identifying the class of the target data through linear superposition of the sampled training data files using l1 minimization, wherein a linear superposition with a sparsest number of coefficients is used to identify the class of the target data.