Age Estimation System Integrating Discrete and Continuous Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing age and gender estimation systems face accuracy issues when handling age groups as discrete quantities, leading to reduced precision in estimating young and middle age groups, and when handling as continuous quantities, there's a tendency for results to be drawn to the average age, making it difficult to accurately estimate ages in specific age groups.
Innovation Solution
An attribute estimation system that integrates discrete and continuous quantity estimation methods, using separate feature extraction units and classifiers for age and gender estimation, with score-generation units to generate scores that are then integrated to improve estimation accuracy, allowing for weighted precision adjustments based on class performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If age is divided into discrete classes (0s, 10s, 20s, 30s, 40s, 50s, 60+), then classification is simplified and specific age groups can be identified, but accuracy declines at class boundaries (e.g., 29-30 years old) and for middle age groups where features are less distinct
Solution Approach 1:
The patent divides the age estimation task into two separate segmentation approaches: discrete classification (dividing age into classes like 0s, 10s, 20s, etc.) and continuous estimation (predicting exact age values). By segmenting the problem into these two independent estimation methods and then integrating their results, the system achieves both classification simplicity and measurement precision across all age groups including boundaries and middle age groups.
2Measurement precision
If age is estimated as a continuous quantity, then precise age values can be obtained, but results tend to be drawn to the center (average age), making it difficult to accurately estimate ages in specific age groups
Solution Approach 1:
The patent merges two different estimation approaches: discrete classification (which provides reliable category identification) and continuous estimation (which provides precise age values). The integration unit combines the discrete class prediction with the continuous age prediction, allowing the system to maintain accuracy in specific age groups while still providing precise age estimates, thereby eliminating the center-drawing tendency of pure continuous methods.
3Productivity
When specific features cannot be extracted from images (as in rising generation and middle age groups), then classification into discrete classes becomes unreliable, but these are precisely the groups where accurate customer base analysis is most needed
Solution Approach 1:
The patent changes the estimation parameters by employing two different parameter representation methods: discrete class labels (0s, 10s, 20s, etc.) and continuous numerical age values. This parameter duality allows the system to handle cases where feature extraction is unreliable by leveraging the complementary strengths of both parameter types, ensuring accurate customer base analysis coverage across all age groups including those with less distinct features.
Data Source
AI summary
An attribute estimation system and a method in which there are no cases that the estimation accuracy declines in a specific numerical value area, and an age estimation system, a gender estimation system and an age and gender estimation system using this is provided.It is a system to estimate an age of a person photographed in an input image, the system including: a classifier 3 that estimates the age of a person as a discrete quantity based on data of an input image; a classifier 4 that estimates the age of a person as a continuous quantity based on data of an input image; and an integration unit 7 that integrates an estimated result of the classifier 3 and an estimated result of the classifier 4.