Multi-matcher fusion for image recognition accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition technologies, particularly deep learning neural networks, face challenges in achieving high accuracy, especially with low-quality images due to resolution issues, pose problems, and illumination, and struggle to maintain accuracy as database searches grow larger.
Innovation Solution
The system and method involve generating a multi-feature multi-matcher fusion (MMF) predictor by training a neural network to extract feature spaces at different resolutions and selecting matcher functions, which are then processed to create weighted feature/matcher components for improved image recognition scoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep learning neural networks are used for image recognition, then recognition capability is improved, but accuracy deteriorates in challenging conditions (low resolution, bad pose, bad illumination)
Solution Approach 1:
The patent segments the feature extraction process by generating feature spaces at multiple resolutions (different levels of detail) rather than using a single fixed resolution. This allows the system to capture both fine-grained and coarse-grained features, improving accuracy in challenging conditions while maintaining general recognition capability.
Solution Approach 2:
The patent creates a composite recognition system by fusing multiple feature spaces (at different resolutions) with multiple matcher functions. This composite approach combines the strengths of different feature representations and matching algorithms to achieve higher accuracy across diverse image quality conditions.
2Quantity of substance
If database searches become large and larger, then coverage is improved, but accuracy deteriorates
Solution Approach 1:
The patent segments the feature space into multiple resolutions, allowing the system to process large databases more effectively by comparing images at different levels of detail. This segmentation enables more precise matching even as database size grows, counteracting the typical degradation in accuracy.
Solution Approach 2:
The patent adds the dimension of resolution levels to the feature space, transforming the problem from a single-dimensional feature comparison to a multi-dimensional approach. This allows the system to maintain accuracy in large databases by selecting appropriate resolution levels for different comparison scenarios.
3Productivity
If fixed number of dimensions (e.g., 256) is used for feature space, then computational efficiency is improved, but accuracy deteriorates in challenging situations
Solution Approach 1:
The patent makes the feature space dynamic by generating features at multiple resolutions rather than using a fixed dimensionality. The system can adaptively select and fuse features from different resolution levels based on the specific comparison needs, improving accuracy while managing computational resources efficiently through the fusion framework.
Solution Approach 2:
The patent changes the parameter of feature space dimensionality from a fixed value to a multi-resolution set. By generating feature spaces at different resolutions and fusing them with weighted matchers, the system achieves higher accuracy in challenging situations while maintaining computational efficiency through the optimized fusion process.
4Device complexity
If single matcher function is used, then system complexity is reduced, but accuracy deteriorates
Solution Approach 1:
The patent merges multiple matcher functions into a unified fusion framework that combines their outputs with learned weights. This merging approach allows the system to leverage the strengths of different matchers (e.g., cosine similarity, Euclidean distance) while presenting a unified interface, improving accuracy without proportionally increasing system complexity.
Solution Approach 2:
The patent creates a composite matching system by combining multiple matcher functions with multiple feature spaces. The fusion mechanism integrates these diverse components into a cohesive system that achieves higher accuracy while managing complexity through the unified fusion architecture.
Data Source
AI summary
A system, method and program product for implementing image recognition. A system is disclosed that includes a training system for generating a multi-feature multi-matcher fusion (MMF) predictor for scoring pairs of images, the training system having: a neural network configurable to extract a set of feature spaces at different resolutions based on a training dataset; and an optimizer that processes the training dataset, extracted feature spaces and a set of matcher functions to generate the MMF predictor having a series of weighted feature/matcher components; and a prediction system that utilizes the MMF predictor to generate a prediction score indicative of a match for a pair of images.