Hierarchical Normalization in Image Descriptor Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image descriptor technologies face challenges in accurately generating and comparing image descriptors to determine similarity between images, particularly in distinguishing between images based on their visual features.
Innovation Solution
A descriptor network is trained to generate image descriptors with major and minor vectors, where hierarchical normalization is applied by normalizing the major vector to a higher amount and the minor vector to a lower amount, allowing for improved performance in image matching tasks by computing distances between these vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If hierarchical normalization is applied with different normalization amounts for major and minor vectors, then image matching accuracy is improved, but computational complexity increases
Solution Approach 1:
The image descriptor is segmented into multiple vectors with different hierarchical levels (first vector, second vector, third vector, etc.). Each vector is normalized independently with different normalization amounts, allowing the system to capture image features at multiple scales and levels of detail, thereby improving matching accuracy while managing computational complexity through structured organization.
Solution Approach 2:
Different normalization amounts are applied to different vectors based on their hierarchical importance. The first vector receives a different normalization amount than the second, third, and subsequent vectors. This local differentiation optimizes the contribution of each vector to the overall descriptor, enhancing matching precision by emphasizing critical features while reducing noise from less important ones.
2Reliability
If multiple vectors with different normalization amounts are used, then robustness of image similarity determination is improved, but the number of processing steps increases
Solution Approach 1:
The image descriptor is pre-processed into multiple normalized vectors before the matching operation. Each vector is independently normalized with appropriate normalization amounts during descriptor generation. This preliminary organization allows for efficient comparison during matching, as the hierarchical structure and normalization are already in place, reducing the computational burden and time required during actual image similarity determination.
Data Source
AI summary
Techniques are disclosed for using and training a descriptor network. An image may be received and provided to the descriptor network. The descriptor network may generate an image descriptor based on the image. The image descriptor may include a set of elements distributed between a major vector comprising a first subset of the set of elements and a minor vector comprising a second subset of the set of elements. The second subset of the set of elements may include more elements than the first subset of the set of elements. A hierarchical normalization may be imposed onto the image descriptor by normalizing the major vector to a major normalization amount and normalizing the minor vector to a minor normalization amount. The minor normalization amount may be less than the major normalization amount.