Linear Texture Model Segmentation for Face Recognition Under Illumination Variations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current statistical models for human faces are sensitive to illumination changes, making it difficult to accurately recognize faces under varying lighting conditions, and they fail to decouple variations in identity from those caused by directional lighting, leading to non-realistic shape/texture configurations and ambiguous parameter interpretation.
Innovation Solution
A linear texture model is constructed with separate subsets for directional lighting variations and those independent of it, allowing for the adjustment of model components to generate a corrected image that is robust to illumination changes, and incorporating eigenvectors and eigenvalues to determine characteristics like features independent of directional lighting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical models are trained using annotated image examples, then the model can represent face appearance, but the model becomes sensitive to illumination changes and fails to distinguish identity variations from lighting variations
Solution Approach 1:
The patent segments the texture space into distinct subspaces: one capturing identity-related variations and another capturing illumination-related variations. By decomposing the overall texture model into these separate components, the system can independently analyze and process identity features without interference from lighting changes, thereby improving both recognition accuracy and robustness
Solution Approach 2:
The patent extracts illumination components from the texture model by identifying and isolating variations that are specifically related to lighting conditions. This extraction process removes the harmful influence of illumination changes from the identity representation, allowing the model to focus on stable identity features while separately accounting for lighting effects
2Device complexity
If the model uses a unified texture representation, then the model structure is simple, but it cannot decouple identity variations from lighting variations leading to ambiguous parameter interpretation
Solution Approach 1:
The patent divides the texture space into multiple specialized subspaces, each responsible for capturing specific types of variations. This segmentation creates clearly defined parameter spaces where identity parameters and illumination parameters are distinct and non-overlapping, eliminating ambiguity in parameter interpretation while maintaining a systematic model structure
3Adaptability or versatility
If the model is trained on images with varying lighting conditions, then the model can handle diverse lighting scenarios, but the variations due to illumination changes dominate over variations between different individuals
Solution Approach 1:
The patent extracts and isolates illumination-related variations from the training data by identifying patterns specific to lighting changes. By removing these dominant illumination components from the identity learning process, the model can train on diverse lighting conditions without allowing lighting variations to overshadow subtle inter-individual differences, thereby maintaining both adaptability and identity discrimination precision
Data Source
AI summary
A technique for determining a characteristic of a face or certain other object within a scene captured in a digital image including acquiring an image and applying a linear texture model that is constructed based on a training data set and that includes a class of objects including a first subset of model components that exhibit a dependency on directional lighting variations and a second subset of model components which are independent of directional lighting variations. A fit of the model to the face or certain other object is obtained including adjusting one or more individual values of one or more of the model components of the linear texture model. Based on the obtained fit of the model to the face or certain other object in the scene, a characteristic of the face or certain other object is determined.