Face Image Classification Using Hybrid Clustering Trigger Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face clustering algorithms require significant recalculation and re-clustering every time a new face feature is added, leading to high computational costs and poor real-time performance.
Innovation Solution
A method that determines the category of a face image by using a neighbor face feature algorithm when the face feature collection does not meet a preset trigger condition, avoiding the need for clustering and reducing computational load, while using full or incremental clustering algorithms based on specific conditions to improve accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full clustering algorithm is used to cluster all face features when a new face feature is added, then classification accuracy is improved, but computational cost increases significantly
Solution Approach 1:
The patent segments the clustering process into two distinct modes: incremental clustering for routine additions and full clustering for periodic comprehensive updates. This segmentation allows the system to use computationally efficient incremental clustering for most operations while periodically applying full clustering to maintain overall accuracy, thus resolving the contradiction between continuous high accuracy and computational cost.
Solution Approach 2:
The patent implements periodic full clustering at predetermined intervals or when trigger conditions are met, rather than performing full clustering every time a new face feature is added. This periodic action reduces computational cost by limiting expensive full clustering operations to necessary moments while maintaining classification accuracy through regular comprehensive updates.
2Use of energy by moving object
If incremental clustering algorithm is used to sample face features, then computational cost is reduced, but classification accuracy deteriorates
Solution Approach 1:
The patent merges incremental clustering and full clustering into a unified hybrid system that leverages the strengths of both approaches. Incremental clustering handles routine operations with low computational cost, while full clustering is periodically applied to correct and refine the incremental results, ensuring classification accuracy is maintained without sacrificing computational efficiency.
Solution Approach 2:
The system uses trigger conditions and predetermined intervals as feedback mechanisms to determine when to switch from incremental to full clustering. This feedback ensures that full clustering is applied when necessary to maintain accuracy, while allowing incremental clustering to operate during periods when accuracy requirements are met by previous full clustering operations.
3Reliability
If face features are re-clustered every time a new face feature is added, then up-to-date classification is achieved, but real-time performance deteriorates
Solution Approach 1:
The patent implements a dynamic clustering strategy that adapts the clustering frequency and type based on system state and trigger conditions. Rather than rigidly re-clustering every time, the system dynamically selects between incremental and full clustering based on whether trigger conditions are met, enabling real-time performance while maintaining up-to-date classification through selective updates.
Solution Approach 2:
The system prepares for potential full clustering operations by monitoring trigger conditions and maintaining readiness to switch modes. This preliminary monitoring and conditional preparation allows the system to quickly transition to full clustering when needed while maintaining real-time performance during normal operations by using efficient incremental clustering.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments of the disclosure provide a method for classifying a face image, belonging to a field of artificial intelligence technologies. A face image is acquired (101). A face features is extracted (102) from the face image. When a face feature collection does not meet a preset trigger condition, a category of the face feature is determined (103) based on categories of existing face features in the face feature collection and a neighbor face feature algorithm. A category of the face image is determined (104) based on the category of the face features.