Data Similarity Matching Device Score Correction Mechanism
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for calculating the similarity between data sets, particularly images, often result in inaccurate matches due to issues like occlusion and varying data amounts, leading to higher similarity scores when less information is used, which can cause incorrect identifications.
Innovation Solution
A matching device and method that selects corresponding elements from first and second data vectors, calculates a similarity score, and corrects it to increase with the amount of data used, ensuring accurate similarity calculations by focusing on usable data areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the position and posture of the three-dimensional shape are adjusted to minimize difference from target image, then the generated image becomes smaller by hiding dissimilar parts, but the matching accuracy deteriorates when comparing with images of different people
Solution Approach 1:
The patent pre-calculates and stores multiple three-dimensional shapes representing the same person under different conditions (occlusion, illumination, posture) before matching. This preliminary preparation allows the system to select the most appropriate pre-computed shape for comparison, rather than generating images on-demand, thereby ensuring reliable matching even when some areas are occluded or dissimilar
Solution Approach 2:
The patent changes the parameters used for similarity calculation by introducing a correction term that accounts for the amount of data used. The similarity score is adjusted based on the ratio of actually used data areas to total data areas, which compensates for situations where occlusion or dissimilarity reduces the effective comparison area
2Ease of operation
If occlusion or dissimilarity hides parts of the three-dimensional shape, then the generated image uses less data area, but the similarity score incorrectly increases
Solution Approach 1:
The patent introduces a feedback mechanism where the calculated similarity score is corrected based on the amount of data actually used for comparison. The correction term feeds back the information about data usage ratio to adjust the final similarity score, preventing incorrect high scores when occlusion or dissimilarity reduces the effective comparison area
Solution Approach 2:
The patent changes the similarity scoring parameter by adding a correction component that considers the ratio of used data areas. This transforms the raw similarity score into a corrected score that accounts for data availability, ensuring that reduced data area due to occlusion does not artificially inflate similarity
3Adaptability or versatility
If multiple three-dimensional shapes of different people are used for matching, then the system can handle variety in data, but incorrect matching occurs when similarity is calculated using reduced data areas
Solution Approach 1:
The patent pre-computes and stores multiple three-dimensional shapes for each person under various conditions (different occlusions, illuminations, postures). This preliminary action creates a comprehensive database that allows the system to handle diverse input images reliably by selecting the best-matching pre-computed shape, preventing incorrect matches even when data areas are reduced
Solution Approach 2:
The patent introduces an intermediary correction mechanism that mediates between the raw similarity calculation and the final matching decision. The correction term acts as an intermediary that adjusts the similarity score based on data usage ratio, ensuring that variety in data handling does not compromise matching accuracy
Data Source
AI summary
Provided is a matching device capable of improving the accuracy of the degree of similarly in the calculation of the degree of similarly between data sets. Element selection unit selects elements corresponding to each other between a first vector including a plurality of elements determined based on first data and a second vector including a plurality of elements determined based on second data. Similarity degree calculation unit calculates a score of the degree of similarly between the first data and the second data from the elements selected from the first vector and the second vector. Score correction unit corrects the score calculated by the similarity degree calculation unit so that the increment of the score increases with the increase in the amount of data used for the calculation of the degree of similarly.