Modality Detection Using Multiple Transformation Matrix Groups
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for modality identification in input data, such as images, face detection, suffer from degraded accuracy due to individual differences, which affects the reliability of object detection.
Innovation Solution
The use of multiple transformation matrix groups, comprising a first projection matrix for projecting input data into a space vector and a second projection matrix for reducing its dimensionality, along with inverse projection and correlation calculation to identify the modality by determining the highest correlation between the original and inverse space vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods map images into feature vector space and detect faces based on distance to subspace, then object detection can be performed, but detection accuracy is degraded due to individual face differences
Solution Approach 1:
The patent segments the detection process into multiple independent projection stages using transformation matrix groups. Each group projects input data through different transformation paths (first projection to space vector, second projection to projection data, then inverse projection back). This segmentation allows the system to evaluate multiple modalities separately and select the most appropriate one, thereby avoiding the accuracy degradation caused by individual differences that affects conventional single-path methods.
Solution Approach 2:
The patent changes the parameter representation by using multiple transformation matrix groups with different projection pathways. Instead of relying on a single feature vector space mapping, the system transforms data through multiple parameter spaces and evaluates correlations between original and transformed representations. This parameter transformation approach enables the system to adapt to different modalities and overcome the limitations of conventional methods that are sensitive to individual differences.
2Measurement precision
If multiple transformation matrix groups are used with inverse projection and correlation calculation, then detection accuracy is significantly enhanced, but device complexity increases
Solution Approach 1:
The complex detection task is segmented into modular transformation matrix groups, where each group handles a specific modality. The system divides the overall transformation process into distinct stages (first projection, second projection, inverse projection) that can be independently computed and evaluated. This modular segmentation makes the complex system more manageable and allows for efficient implementation despite the multiple transformation paths.
Solution Approach 2:
The patent applies partial action by using multiple transformation matrix groups, but not all possible transformations need to be fully executed for every input. The system performs projections and inverse projections selectively based on the input data characteristics, and uses correlation calculation to identify the most relevant modality. This approach avoids the need to fully process all transformation paths, reducing the effective computational complexity while maintaining high detection accuracy.
3Productivity
If conventional methods use single subspace projection, then processing is computationally efficient, but individual differences cause accuracy degradation
Solution Approach 1:
The patent segments the processing into multiple transformation matrix groups that can be efficiently computed using standardized projection operations. Each segment (projection to space vector, projection to projection data, inverse projection) uses optimized linear algebra operations that maintain computational efficiency. The segmentation allows parallel processing of different transformation paths and enables the system to quickly evaluate multiple modalities without sacrificing overall processing speed.
Solution Approach 2:
The transformation matrix groups are pre-computed and stored, allowing the system to perform rapid projections during actual detection. The preliminary preparation of transformation matrices enables the system to efficiently process input data through multiple pathways without the computational overhead of calculating transformations in real-time. This preliminary action maintains processing efficiency while enabling the use of multiple transformation paths for improved accuracy.
Data Source
AI summary
Disclosed is a method for accurately and efficiently detecting a modality of input data, including the steps of projecting the input data into a plurality of projection data using each of a plurality of transformation matrix groups U1·(Σ12U2T), generating a plurality of inverse projection data by performing inverse projection of the transformation matrix groups on the plurality of generated projection data, calculating a correlation between the input data and the generated inverse projection data with respect to each transformation matrix group U1·(Σ12U2T), and identifying a modality represented by a transformation matrix group having a highest calculated correlation as the modality of the input data.