IF-Net Image Processing for Lighting Invariant Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face challenges in maintaining feature invariance and uniqueness under varying outdoor lighting conditions and viewing angles, leading to suboptimal matching results due to low tolerance for lighting interference and field of view variations.
Innovation Solution
An image processing method and system that utilize a feature extraction model based on IF-Net deep learning structures, adjust the model by selecting and calculating loss function values from training blocks to improve feature descriptor adaptability, and update net parameters to enhance matching proficiency under changing lighting conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional feature extraction methods are used, then the system is simple and easy to implement, but the matching proficiency deteriorates under varying outdoor lighting conditions
Solution Approach 1:
The patent implements dynamic adaptation by training the feature extraction model with multiple loss functions that adjust to different lighting conditions. The model dynamically updates its parameters based on outdoor lighting variations, transforming a static feature extraction system into a dynamic one that adapts to changing environmental conditions, thereby improving matching proficiency under varying lighting.
Solution Approach 2:
The patent changes the parameters of the feature extraction model by introducing multiple loss functions (classification loss, feature loss, and light interference loss) that modify the model's weight parameters during training. This parameter adjustment enables the model to better distinguish features under different lighting conditions, resolving the contradiction between reliability and complexity.
2Measurement precision
If feature invariance is maintained under different viewing angles, then orientation accuracy improves, but the tolerance for lighting interference worsens
Solution Approach 1:
The patent introduces a light interference loss function as an intermediary mechanism that mediates between feature invariance and lighting interference tolerance. This loss function specifically addresses lighting variations by adding a dedicated training objective that reduces the model's sensitivity to lighting changes while maintaining orientation accuracy, effectively decoupling these two previously conflicting requirements.
Solution Approach 2:
The patent applies local quality by differentiating the treatment of different feature aspects: orientation features are optimized for invariance under viewing angle changes, while appearance features are optimized for robustness against lighting interference through separate loss function components. This localized optimization resolves the contradiction by allowing different parts of the feature space to have different properties.
3Reliability
If the feature extraction model is adjusted with multiple loss functions, then matching proficiency under changing lighting conditions improves, but the training complexity and calculation time increase
Solution Approach 1:
The patent applies partial action by selectively applying different loss functions based on the specific training phase and data characteristics. Rather than always using all loss functions at full strength, the training process strategically combines classification loss, feature loss, and light interference loss in a phased manner, reducing overall training time while maintaining the benefits of multiple loss functions for improving matching proficiency.
Data Source
AI summary
An image processing method includes analyzing multiple images data based on Illumination-invariant Feature Network (IF-NET) with an image processing device to generate corresponding sets of eigenvector, in which image data includes a first image data related to at least one first feature of the sets of eigenvector, and a second image data related to at least one second feature of the sets of eigenvector; choosing a corresponding first training set of tiles and second training set of tiles from the first image data and second image data with an image processing device based on IF-NET, and computing on both training set of tiles to generate a least one loss value; and adjusting IF-NET based on a least one loss value. An image processing system is also disclosed herein.