Scale Robust Feature-Based Image Identifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image identification methods are not robust enough to handle significant modifications such as cropping and translation, leading to reduced detection rates and increased computational costs, while also requiring high levels of precision and speed in large datasets.
Innovation Solution
A method that derives an image descriptor by processing scale-space representations to detect feature points, selecting scale-dependent image regions, and using a four-stage matching process with the Trace transform to maintain robustness while minimizing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If prior art feature point methods extract large numbers of features to densely cover the image, then detection coverage is improved, but searching and matching calculations become very slow
Solution Approach 1:
The patent divides the image into multiple scale levels using scale-space representation, and at each scale level, only a limited number of feature points are extracted. This segmentation approach distributes the detection coverage across multiple scales rather than requiring all features at a single scale, thereby maintaining comprehensive coverage while reducing the number of features that need to be processed simultaneously, which improves searching and matching speed.
Solution Approach 2:
The patent introduces the scale dimension by constructing a scale-space representation of the image. Instead of extracting features only at the original image scale, features are extracted across multiple scales. This additional dimensional approach allows the system to achieve comprehensive detection coverage through multi-scale feature points while keeping the number of features at each scale manageable, thus improving both detection coverage and processing efficiency.
2Productivity
If prior art methods use various forms of data-structures to reduce search times, then searching speed is improved, but additional database initialisation times, additional storage and memory requirements increase
Solution Approach 1:
The patent extracts only the essential feature information needed for matching by deriving scale-dependent image region representations from the scale-space feature points. Rather than storing and processing complete images or complex data structures, the method extracts and stores only the critical regional representations, which reduces storage requirements and simplifies the data structures while maintaining searching speed through efficient feature-based comparison.
3Reliability
If prior art identifiers are designed for high robustness to cropping and translation, then robustness is improved, but discrimination ability and false alarm rates worsen
Solution Approach 1:
The patent derives scale-dependent image region representations centered at each detected feature point, rather than using global image features. Each local region is processed and represented independently, allowing the system to maintain robustness to global transformations like cropping and translation while preserving local discriminative information. The local regional representations retain sufficient detail to distinguish between different images even when globally modified.
Data Source
AI summary
A method for deriving an image identifier comprises deriving a scale-space representation of an image, and processing the scale-space representation to detect a plurality of feature points having values that are maxima or minima. A representation is derived for a scale-dependent image region associated with one or more of the detected plurality of feature points. In an embodiment, the size of the image region is dependent on the scale associated with the corresponding feature point. An image identifier is derived using the representations derived for the scale-dependent image regions. The image identifiers may be used in a method for comparing images.