Spiral Aggregation Map for Rotation-Invariant Image Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image description methods, such as ring and radial projection transformations, face limitations in handling circular symmetry and rotation, leading to information loss and inefficiencies in image matching, especially with gray scale images which have less information compared to color images.
Innovation Solution
The method employs a spiral aggregation track to create a spiral aggregation map, allowing for image rotation, size change resistance, and angle variation detection by sampling pixels along a spiral trajectory and ranking their values, forming a descriptive model that identifies image objects through multiple steps and comparisons.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ring projection transformation is used, then anti-rotation capability is improved, but applicability to circular symmetry and radial patterns deteriorates
Solution Approach 1:
The patent segments the image description into multiple spiral aggregation tracks at different angles (0°, 45°, 90°, 135°), each track independently capturing radial patterns from the center outward. This segmentation allows the system to maintain anti-rotation capability while simultaneously capturing circular symmetry and radial patterns through multiple oriented views.
Solution Approach 2:
The patent introduces a new dimensional approach by creating spiral aggregation tracks that extend radially from the center to the periphery, rather than using conventional ring-based circular transformations. This radial-spiral dimensionality enables simultaneous capture of rotation, circular symmetry, and radial patterns in a unified representation.
2Adaptability or versatility
If radial projection transformation is used, then circular symmetry and radial patterns are captured, but anti-rotation capability deteriorates
Solution Approach 1:
The patent divides the radial projection into multiple segmented spiral aggregation tracks, each oriented at specific angles (0°, 45°, 90°, 135°). Each track independently captures radial patterns while the collection of tracks together provides rotation invariance, resolving the contradiction between circular symmetry capture and anti-rotation capability.
Solution Approach 2:
The patent merges multiple radial projection transformations at different angles into a unified spiral aggregation map. This combination integrates the circular symmetry capture capability of radial projection with the anti-rotation capability by aggregating information from multiple oriented views.
3Device complexity
If conventional image description methods are used, then processing is simple, but information loss occurs during rotation and size change
Solution Approach 1:
The patent performs preliminary spiral aggregation transformation on the image before recognition, creating a rotation-invariant representation in advance. This preliminary action preserves structural and continuous information during subsequent processing, preventing information loss during rotation and size change while maintaining reasonable processing complexity.
Solution Approach 2:
The patent creates multiple copies of the image data along spiral aggregation tracks at different angles and radial positions. These copies preserve the original image information in multiple transformed domains, preventing information loss during rotation and size changes while maintaining processing efficiency through the copied representations.
Data Source
AI summary
Image description and image recognizable method, it contain (a) It obtain an image which possess plural pixels. (b) It determines a starting position in the image. (c) In the image, From the starting point along the trajectory of the former spiral aggregation makes a pixel sampling, and the pixel on the trajectory rank to the former spiral aggregation. (d) the angle increases with the increase of the variance, it forming a the angle of the latter spiral aggregation. From the starting point along a trajectory of the former spiral aggregation makes the pixel sample, and the pixel on the trajectory rank to the former spiral aggregation. (e) It decides how many frequencies the angle variation increase, and repeatedly performs the step (d). After obtaining a plurality of the latter spiral aggregation, the pixel corresponds to the value. (f) It ranks the former spiral aggregation and the latter spiral aggregation. Then, spiral aggregation map will be formed and recorded the every value of the pixel.