Weighted Likelihood Integration for Human Age Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing age estimation methods using one-dimensional attribute values face inaccuracies due to high likelihood outputs for both correct and incorrect age classes, especially when facial features indicate different ages, leading to estimated ages being far from the correct age.
Innovation Solution
An estimating apparatus that acquires images, extracts human features, calculates first and second likelihoods for consecutive attribute classes, and integrates these likelihoods to correct the estimated attribute value by applying weights to selected classes, thereby improving the accuracy of age estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If all results of determination from multiple age class determinations are integrated to estimate age, then the estimation process can be completed, but the estimated age may be far away from the correct age when facial features show conflicting signals
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different age class determinations based on their spatial proximity in the attribute space. Determinations from adjacent age classes are given higher weight than those from distant classes, creating a localized quality assessment that prioritizes more relevant evidence while reducing the impact of irrelevant conflicting signals.
Solution Approach 2:
The patent changes the parameter of likelihood weighting by introducing a weight generation mechanism that adjusts the influence of different determinations based on their proximity to the target age class. This parameter change transforms the simple integration process into a weighted integration process, where the weight itself is determined by the distance metric in attribute space.
2Adaptability or versatility
If multiple determiners are used to cover different age ranges, then the coverage of age estimation is improved, but the likelihood of incorrect age classes increases when facial features are ambiguous
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different age class determinations based on their spatial proximity in the attribute space. Determinations from adjacent age classes are given higher weight than those from distant classes, creating a localized quality assessment that prioritizes more relevant evidence while reducing the impact of irrelevant conflicting signals.
Solution Approach 2:
The patent introduces feedback by using the distance metric to dynamically adjust the weights of determinations based on their consistency with the target age class. The distance calculation provides feedback about the reliability of each determination, allowing the system to automatically down-weight unreliable determinations from ambiguous facial features.
Data Source
AI summary
An estimating apparatus configured to estimate a correct attribute value is provided. The estimating apparatus extracts feature quantities from an image including a person, calculates a first likelihood of the feature quantity for respective attribute classes; calculating second likelihoods for the respective attribute classes from the first likelihoods for the respective attribute classes; specifies the attribute class having the highest second likelihood; calculates an estimated attribute value of the specific attribute class and estimated attribute values of selected classes by using the feature quantity; and applies the second likelihood on the estimated attribute value of the specific attribute class as a weight, applies the second likelihoods on the estimated attribute values of the selected classes as a weight and add the same, and calculates a corrected attribute value of the specific attribute class.