Image Feature Extraction via Multi-Resolution Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image matching techniques in computer vision face challenges in mobile environments due to factors like size, lighting, obstacles, and motion, which affect the precision of feature recognition, especially in low-light conditions.
Innovation Solution
A method involving the estimation of initial key points from an input image, generation of descriptors based on a downscaled image, and matching these points to obtain feature points, which improves robustness against noise and enhances singularity, enabling accurate correspondence point estimation between images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image matching is performed using original resolution images, then feature recognition precision is improved, but processing time and computational complexity increase
Solution Approach 1:
The patent divides the image processing into multiple resolution levels: first processing a downsampled version of the image to obtain initial feature points, then processing the original resolution image to refine these points. This multi-scale segmentation approach reduces overall processing time while maintaining feature recognition precision by focusing computational resources strategically at different scales.
Solution Approach 2:
The patent performs preliminary processing on a downsampled image to identify candidate feature points before processing the full-resolution image. This preliminary action at lower resolution reduces the number of points requiring intensive processing at higher resolution, thereby reducing total processing time while preserving feature accuracy.
2Quantity of substance
If feature detection is performed on images with large motion or in dark environments, then comprehensive feature coverage is improved, but detection accuracy deteriorates
Solution Approach 1:
The patent changes the resolution parameter by processing images at multiple scales. By first analyzing a downsampled version of the image, the system can detect features in dark or high-motion conditions where details are less critical, then use these results to guide more precise detection at original resolution, thereby maintaining accuracy while achieving comprehensive coverage.
Solution Approach 2:
The patent adds the dimension of scale by introducing multi-resolution processing. Instead of attempting to detect all features at full resolution simultaneously (which fails in difficult conditions), the system operates across multiple resolution dimensions, using coarse-scale detection to guide fine-scale detection, thus achieving both comprehensive coverage and maintained accuracy.
3Measurement precision
If descriptors are generated from original resolution images, then descriptor accuracy is improved, but computational complexity and memory usage increase
Solution Approach 1:
The patent segments the descriptor generation process into two stages: first generating descriptors from a downsampled image to identify candidate feature points, then generating refined descriptors only for these candidate points from the original resolution image. This segmentation reduces the total number of high-complexity operations while preserving descriptor accuracy for the final output.
Solution Approach 2:
The patent applies partial action by generating full-precision descriptors only for a subset of candidate feature points identified from the downsampled image, rather than processing all possible points at full resolution. This partial processing approach reduces computational complexity and memory usage while maintaining sufficient accuracy for the matched features.
Data Source
AI summary
In a method of extracting features from an image, a plurality of initial key points are estimated based on an input image. A plurality of descriptors are generated based on a downscaled image that is generated by downscaling the input image. A plurality of feature points are obtained by matching the plurality of initial key points with the plurality of descriptors, respectively.