Robust Signatures via Multi-Axis Nonlinear Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Video fingerprinting techniques, such as ordinal and differential signatures, are susceptible to changes like cropping, rotation, and geometric transformations, making them ineffective for robust content recognition.
Innovation Solution
The use of multi-axis filtering operations to derive signatures based on mean subblock luminances, which are more robust to local changes and geometric transformations, and can be adapted for rotation, using nonlinear filters like Octaxis and Gaussian weighting for improved resilience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ordinal or differential signatures are used for video fingerprinting, then content recognition can be performed, but the signatures become susceptible to geometric transformations, cropping, and local changes
Solution Approach 1:
The patent applies local nonlinear filtering operations (such as multi-axis comparison filters) to compute local signatures at different positions in the video frame. Each local signature is derived from a specific region's luminance relationships, allowing the system to maintain robustness against global transformations while being sensitive to local changes. This local quality approach enables the system to handle cropping and geometric transformations by focusing on invariant local patterns rather than global ordering.
Solution Approach 2:
The patent transforms the signature computation from using global ordinal rankings or differential values to using local nonlinear filter outputs. By changing the parameter space from global luminance ordering to local multi-axis filtered values, the system achieves invariance to geometric transformations and cropping while maintaining ability to detect local modifications like logo insertion or subtitles.
2Reliability
If multi-axis filtering operations are applied to derive robust signatures, then robustness to local changes and geometric transformations is improved, but computational complexity increases
Solution Approach 1:
The patent divides the video frame into multiple local regions or subblocks and applies the multi-axis filtering operation independently to each region. This segmentation allows the complex filtering to be performed in parallel across multiple simple units, reducing the overall computational burden while maintaining the robustness benefits. Each local signature computation is a simplified version of the full filtering operation applied only to its specific region.
Data Source
AI summary
Content signal recognition is based on a multi-axis filtering of the content signal. The signatures are calculated, formed into data structures and organized in a database for quick searching and matching operations used in content recognition. For content recognition, signals are sampled and transformed into signatures using the multi axis filter. The database is searched to recognize the signals as part of a content item in the database. Using the content identification, content metadata is retrieved and provided for a variety of applications. In one application, the metadata is provided in response to a content identification request.