Augmented Image Correlation for Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Object recognition in images is hindered by dynamic environmental changes such as varying lighting, background, and orientation, leading to inconsistent correlation scores between objects and templates, making it difficult to accurately identify objects.
Innovation Solution
Augmenting traditional image correlation techniques by extracting and matching object features like corners, edges, and regions of interest from source images to corresponding feature templates, which enhances the correlation score and improves recognition accuracy under adverse conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image correlation techniques are used, then the matching process is simple, but the recognition accuracy deteriorates under changing environmental conditions
Solution Approach 1:
The patent segments the object recognition process into two independent correlation operations: traditional pixel-based image correlation and feature-based correlation. Features such as corners, edges, and ridges are extracted from both the object and template, then correlated separately. The final recognition decision combines results from both correlation methods, allowing the system to maintain accuracy under varying environmental conditions while managing complexity through modular processing.
2Measurement precision
If multiple templates with different scales and rotations are used, then the matching accuracy improves, but the computational complexity and time consumption increase significantly
Solution Approach 1:
The patent transforms the template matching problem from spatial domain to feature domain by extracting invariant features (corners, edges, ridges) that remain consistent across different scales and rotations. Instead of generating and comparing multiple transformed templates, the system extracts features from a single template and compares them with features extracted from the object at any orientation or scale, dramatically reducing computational time while maintaining matching accuracy.
3Reliability
If feature extraction and matching is added to traditional correlation, then the recognition reliability improves, but the processing complexity increases
Solution Approach 1:
The patent merges traditional pixel-based image correlation with feature-based correlation into a unified recognition framework. Both correlation methods process the same object-template pair independently, then their results are combined to make the final recognition decision. This combination leverages the strengths of both approaches: traditional correlation captures overall appearance while feature correlation provides robustness to environmental variations, thereby improving reliability without requiring completely separate systems.
Data Source
AI summary
Systems and methods search pixels of source images and compare them to pixels of templates. Best matches correlate to objects in the image. That environmental conditions impact the appearance of objects, best matching scores suffer under poor lighting and other adverse conditions. Improving scores includes augmenting traditional correlation techniques with object features extracted from the source image and matching them to templates corresponding to the features. Certain embodiments contemplate corrupting pixels of image templates corresponding to objects with pixels extracted from the source image corresponding to features. Representative features include corners, edges, ridges, points/regions of interest, etc. Other embodiments note augmented correlation as a computing application and computing devices therefore, including cameras for capturing images and displaying results to users.