Pattern Recognition Using Local Correlation Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition techniques are affected by variations in illumination, face orientation, and expression, leading to reduced recognition accuracy and increased processing costs due to the need for inverse transformation processes and similarity integration methods that can be flawed.
Innovation Solution
A pattern recognition method and apparatus that calculates correlation values of feature quantities between input and dictionary data, integrates these values to determine similarity, and identifies data attributes based on integrated similarity, thereby reducing the impact of variations and processing costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If inverse transformation processes are executed to remove all predicted variation factors from the input image, then recognition accuracy is improved, but processing cost increases
Solution Approach 1:
The face image is divided into multiple local areas, and correlation values are calculated for each local area separately. This segmentation allows the system to process only relevant portions of the image rather than applying complex inverse transformations to the entire image, thereby maintaining recognition accuracy while reducing processing cost.
Solution Approach 2:
The patent extracts correlation values from corresponding local areas between the input image and dictionary images. By taking out only the essential feature quantities (correlation values) from local areas rather than processing entire images through inverse transformations, the system achieves accurate recognition with lower processing costs.
2Reliability
If inverse transformers are used to remove variation factors, then robustness to variations is improved, but feature quantities indicating individual differences may be deleted
Solution Approach 1:
Different local areas of the face are treated with different correlation value calculation approaches. By focusing on local areas that contain discriminative features for individual identification while being less sensitive to variation factors, the system maintains robustness to variations without deleting individual difference information.
Solution Approach 2:
The patent changes the approach from inverse transformation (which removes variations) to correlation value calculation (which measures similarity). By using correlation values as the parameter for comparison, the system achieves robustness to variations while preserving individual difference information inherent in the correlation measurements.
3Speed
If similarity integration is performed using threshold processes, then recognition speed is improved, but recognition accuracy may be reduced when local areas have similar variations
Solution Approach 1:
Instead of using a fixed threshold for similarity integration, the patent dynamically determines thresholds based on the distribution of correlation values. This dynamic approach allows the system to adapt to different variation conditions, maintaining both recognition speed through threshold-based filtering and accuracy by adjusting thresholds according to the specific input image characteristics.
Data Source
AI summary
A pattern recognition apparatus that recognizes a data attribute of input data calculates correlation values of feature quantities of corresponding local patterns between the input data and dictionary data for each of a plurality of dictionary data prepared for each data attribute, combines, for each data attribute, the calculated correlation values of local patterns of each dictionary datum to acquire a set of correlation values of each data attribute, integrates correlation values included in each set of correlation values of each data attribute to calculate a similarity of the input data for each data attribute, and identifies the data attribute of the input data based on the calculated similarity.