Image Signature Extraction via Sub-region Shape Variation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image signature extraction methods have low discrimination capability due to high correlation between local regions in images and inability to detect frequency components with cycles matching the rectangle's dimensions, leading to reduced robustness and discrimination performance, especially in images with repeated patterns or specific textures.
Innovation Solution
An image signature extraction device that extracts region features from sub-regions with varying shapes and relative positions, generating an image signature based on these features to improve discrimination capability by reducing correlation between dimensions and enhancing robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If the same shape of local regions is used for extracting features in each dimension, then the extraction process is simple and consistent, but the correlation between dimensions becomes large, reducing discrimination capability
Solution Approach 1:
The patent applies local quality by assigning different shapes to local regions (sub-regions) in different dimensions of the feature vector. Specifically, in at least one dimension, the sub-regions have shapes that differ from those in other dimensions. This makes each dimension extract features with different local characteristics, reducing correlation between dimensions and improving discrimination capability while maintaining extraction simplicity through systematic shape variation.
Solution Approach 2:
The patent employs asymmetry by using non-uniform shapes for sub-regions across different dimensions. Instead of applying the same symmetric rectangular shape to all dimensions, the patent introduces asymmetric shape variations in at least one dimension, which helps break the correlation structure and enhances the discrimination capability of the image signature.
2Ease of manufacture
If rectangle regions are used for feature extraction, then the extraction method is simple and robust, but frequency components with cycles matching the rectangle dimensions cannot be detected, creating blind spots
Solution Approach 1:
The patent applies parameter changes by varying the shape parameters (width, height, aspect ratio) of sub-regions across different dimensions. This prevents the extraction method from being sensitive to specific frequency cycles that would create blind spots with uniform rectangles, while maintaining extraction simplicity through systematic parameter variation. The shape parameters are changed such that at least one dimension has sub-regions with different shapes, eliminating frequency detection blind spots.
3Ease of operation
If features are extracted from images with high correlation between local regions, then the extraction process is straightforward, but the discrimination capability is significantly reduced
Solution Approach 1:
The patent addresses high correlation between local regions by extracting features from sub-regions with different shapes in different dimensions. This local quality variation ensures that even when the overall image has high correlation, the feature extraction process captures diverse local characteristics, reducing the impact of high correlation and maintaining discrimination capability while keeping the extraction process straightforward.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The image signature extraction device includes an extraction unit and a generation unit. The extraction unit extracts region features from respective sub-regions in an image in accordance with a plurality of pairs of sub-regions in the image, the pairs of sub-regions including at least one pair of sub-regions in which both a combination of shapes of two sub-regions of the pair and a relative position between the two sub-regions of the pair differ from those of at least one of other pairs of sub-regions, and being classified into a plurality of types based on a combination of shapes of two sub-regions and a relative position between the two sub-regions of each of the pairs. The generation unit generates an image signature to be used for identifying the image based on the extracted region features of the respective sub-regions.